KR20140138784A - Improved dash client and receiver with playback rate selection - Google Patents

Abstract

The client device provides streaming media and includes source components for coupling to the stream manager, the request accelerator, the stream manager, and the request accelerator and for determining which requests to receive. The rate selection process may make rate determinations so that when the buffer is late, the buffer is populated so as to avoid irregularly changing rates and to quickly select an accurate stable rate. Multimedia download for HTTP that allows accurate rate estimates, achieves link capacity even with high network delays and packet loss rates, achieves timely delivery of streams, and achieves relatively stable download rates with near-term variability Strategies can be used. The receiver can use multiple HTTP connections, decompose media requests into smaller chunked requests, synchronize connections using TCP flow control mechanisms, and request data in bursts. The receiver can also use an HTTP pipelining process to keep the connection busy.

Description

[0001] IMPROVED DASH CLIENT AND RECEIVER WITH PLAYBACK RATE SELECTION [0002]

Cross-references to related applications

This application claims priority to U. S. Provisional Application No. 61 / 603,569, filed February 27, 2012, entitled " Improved DASH Client and Receiver with Rate Adaptation and Downloading for Adaptive Video, , The entirety of which is hereby incorporated by reference in its entirety.

DASH refers to " Dynamic Adaptive Streaming over HTTP ". With DASH, a content provider formats content into segments, fragments, representations, adaptations, and so on, along with associated metadata such as MPD files, and stores them all in a standard HTTP server or special HTTP And stores them as files available through the server. A DASH client is a receiver that obtains these files needed to provide a presentation to a user of a DASH client.

DASH clients have severe constraints because users typically want high-quality streaming in a network-constrained environment without warning. Thus, improved DASH clients are desirable.

The client device includes a stream manager for providing streaming media, a stream manager for controlling streams, a request accelerator for receiving network requests for content, a stream manager and a source component for determining which requests to be coupled to the stream accelerator, And a media player. The request accelerator includes a request data buffer for buffering the requests and logic for returning complete responses to each request that can be answered. The stream manager, the request accelerator, and the source component may be implemented as processor instructions or program code, and the client device may further include a program memory, a working memory, a processor, and a power source. The client device may also include a display and a user input device. Client tasks may be parsed between the source component, the stream manager, and the request accelerator to efficiently stream data.

As described herein, in various aspects, a client may perform operations such as determining when to maintain a presentation or switch to another presentation, determine which fragments to request, , It can be ensured that sufficient data can be obtained to continue the stream without stalling.

The rate selection process utilizes a buffer to avoid changing the rates abnormally, (a) when the buffer is at a low level, the buffer is filled, (b) even if low download rate estimates are observed, (c) In a stable rate scenario, rate decisions can be made to quickly select the right stable rate. (a) allow accurate estimation, (b) achieve link capacity even at high network delays and packet loss rates, (c) achieve timely delivery of the stream, and (d) Multimedia download strategies are used for HTTP to achieve stable download rates. To achieve this, the receiver uses a number of HTTP connections, decomposes media requests into smaller chunked requests, synchronizes connections using TCP flow control mechanisms, and requests data in bursts . The receiver can also use an HTTP pipelining process to keep the connection busy.

In a receiver that receives media to play out using a presentation element of the receiver, the playout results in media being consumed at a playback rate from a presentation buffer, and the receiver is configured to select from a plurality of playback rates And the method for selecting the playback rate comprises the step of the receiver monitoring the presentation buffer, wherein the presentation buffer includes at least a time at which the media data is received and a time at which the media data is transmitted to the presentation element associated with the receiver And stores the media data between the times consumed by the user. The receiver also stores an indication of the buffer level - the buffer level corresponds to how much of the presentation buffer has been occupied by media data that has been received but has not yet been consumed by the presentation element - And uses a stored indication and an estimated download rate to calculate a target playback rate and selects from among the plurality of playback rates according to the target playback rate.

The selected playback rate is a playback rate that is less than or equal to a predetermined multiplication of the estimated download rate, and the predetermined multiplication is an increasing function of the buffer level. The predetermined multiplication may be an affine linear function of the playback duration of the media data of the presentation buffer and / or may be less than one when the buffer level of the presentation buffer is less than the threshold amount, and / May be greater than or equal to 1 when the presentation duration of the media data is greater than or equal to a preset maximum amount of presentation time. The predetermined multiplication may be a piecewise linear function of the playback duration of the media data of the presentation buffer. The selected playback rate may be a playback rate that is less than or equal to a predetermined multiplication of the estimated download rate and the predetermined multiplication may be a function of increasing the number of bytes of media data in the presentation buffer. The playback rate is the largest available play rate among a plurality of playback rates that is less than or equal to the proportional factor multiplied by the estimated download rate, and the proportional factor is the media of the presentation buffer divided by the estimate of the response time to rate changes It is an increasing function of playback duration of data.

The estimation of the response time may be an upper limit on the presentation time between switch points of the media data and / or may be an average of the presentation time between switch points of media data and / or a predetermined constant multiplied by an estimated round - may be greater than or equal to the trip time ("ERTT").

The receiver may also determine a permissible deviation of the buffer level, calculate a target playback rate using the stored indication of the buffer level and the allowed deviation of the buffer level, and determine, based on the target playback rate, You can choose.

The playback rate may be selected based on an estimate of the response rate for high proportional factors, low proportional factors, download rate estimates, current playback rate, buffer level, and rate changes. The high proportional and low proportional factors may all be incremental functions of the playback duration of the media data of the presentation buffer divided by an estimate of the response time to the rate change and / or delineated linear functions. The high proportional factor may be greater than or equal to the low proportional factor. The playback rate may be the same as the previous playback rate if the previous playback rate is between a low proportional factor multiplied estimated download rate and a high proportional factor multiplied download rate estimate. If the previous playback rate exceeds the high proportional factor times the estimated download rate, then the playback rate can be selected to be the highest available playback rate that is not higher than the estimated rate multiplied by the proportional factor, and / or the previous playback rate If it is below the low proportional factor multiplication download rate estimate, it can be selected to be the largest available play rate that is not greater than the low proportional factor times the download rate estimate.

In one embodiment, the receiver is configured to receive media to play out using the presentation elements of the receiver, consume media data at a playback rate, and provide a presentation interface to provide playback at one of a plurality of playback rates A presentation buffer that stores media data between at least the time at which the media data is received and the time at which the media data is consumed by the presentation element associated with the receiver and is coupled to the presentation interface, The storage-buffer level for variables regarding buffer capacity corresponds to how much of the presentation buffer is occupied by media data that has been received but has not yet been consumed by the presentation element - , And using the target playback display is stored to calculate the rate, and the estimated download rate, includes logic for arranging the request according to the determined selected playback rate.

Various elements may be implemented using a computer readable medium for execution by a processor for controlling data download via a network path between a source and a receiver coupled by a network path. The computer readable medium may be a non-transitory computer readable medium.

Other aspects of the invention will become apparent from these descriptions.

Figure 1 shows various elements including a DASH client in a DASH deployment and shows how media records arrive at the end user with recording, content preparation and content delivery steps.
2 shows an exemplary architecture of a DASH client with other components, including a stream manager, a request accelerator, a source component, a network connection, and a media player.
Figure 3 is a timing chart of the presentation switching processes and includes Figure 3A for the backward looking process and Figure 3B for the forward looking process.
4 is a timing chart showing a presentation switching process in the case where the switch points are aligned.
5 is a chart showing rates during a time that is managed by a rate estimator, particularly a buffer level adaptive ( pker -type rate estimator).
Figure 6 is a chart showing rate increase versus download time (r-time) when a non-adaptive exponential weighted moving average (" EWMA ") filter is used.
7 is a chart showing rate increase vs. playback time (p-time) when a non-adaptive EWMA filter is used.
8 is a chart showing rate increase versus download time (r-time) when a variable window size weighted shift (" WMA ") filter is used.
Figure 9 is a chart showing the rate increase vs. playback time (p-time) when a pker -type process is used.
10 is a chart showing rate reduction versus download time when the pker process of section 2.1 is used.
Figure 11 illustrates the operation of the pker process for sudden increases in rate.
Figure 12 illustrates the operation of the pker process for sudden drops in rate.
Figure 13 shows a comparison of a simple (fixed-width) moving window average and an exponentially weighted moving average.
14 is a flowchart of a pker rate estimation process.
FIG. 15 shows how the B and T _fast values used by the pker process can be determined from the recorded (Tp, Tr) values in conjunction with FIG.
Figure 16 illustrates aspects of determining values.
Fig. 17 shows the operation of the " watermark " fetching process.
Figure 18 shows examples of lambda and mu functions that can be used to select the playback rate.
Figure 19 shows an exemplary selection of (lambda, mu) -functions using a " conservative " setting.
Figure 20 shows an exemplary selection of (lambda, mu) -functions using a " moderate " setting.
Figure 21 shows an exemplary selection of (lambda, mu) - functions using an " offensive " setting.
Figure 22 shows an exemplary selection of (lambda, mu) - functions using a process for emulating an MLB process to some extent.
Figure 23 shows an example of side-by-side values for lambda settings.
24 shows an example of side by side values for mu settings.
25 shows a process for rate estimation, then rate-based rate selection, and then buffer management-based rate selection.
Fig. 26 shows the rate drop without request cancellation.
FIG. 27 shows a rate drop with a request cancellation.
28 is a flowchart showing an exemplary request canceling process.
Fig. 29 shows a process for request cancel detection.
30 is a diagram of operations that fetch with multiple TCP connections but without receive buffer tuning.
31 is a diagram of other operations that fetch into a plurality of TCP connections and have receive buffer tuning.
32 is a flowchart of an exemplary request accelerator process.
33 shows a process for finding a plurality of sub-requests to be made for a predetermined fragment request.
34 shows a process for selecting individual requests to be selected as the disassembly periods of the source requests having the calculated sizes.
35 shows an example of a fragment structure for a waterline segment determined by time offsets and time offsets.
Figure 36 includes tables of values that may be used for lambda and mu in rate selection.

As shown in FIG. 2, the DASH client described herein includes a stream manager (SM), a request accelerator (RA), a source component (SC), a network connection, and a media player. The DASH client may also include one or more media data buffers. In some embodiments, the RA, SC, and media player all may have their own data buffers, or logical partitions of one large data buffer. In other embodiments, the media player may use all of the data buffers set by the SC, with only the RA having a data buffer for buffering the requests so that the RA can return a complete response for all requests that it can respond to. The SM may have its own (physical or logical) local storage to store the metadata needed to make those decisions.

Figure 1 shows a DASH deployment, with a DASH client.

Figure 2 shows an exemplary architecture of a DASH client with other components. It should be understood that the SM, RA, SC, and media player may be implemented in hardware, software, or some combination. Thus, when a function is assigned to a component, it may be implemented as processor instructions, program code, or the like, and in such cases (such as program memory, ROM, RAM, processor, power supply, connectors, , Etc.), the hardware necessary to execute such instructions is implied. When the network functions are described, it should be understood that a network connection exists and may be wired, optical, wireless, etc., Etc.) User interface functions are also implied.

The DASH client maintains two clocks, or their logical equivalents. One clock is a soft circuit or a real time clock circuit that displays the time of the local clock running on the client, and the other clock is the presentation time, which indicates the presentation time of the media content in relation to its start. Here, the real-time clock time is referred to as "r-time" and the "p-time" is a descriptor for presentation time.

The representations are media streams encoded at different bit-rates or other differences for the same content. Thus, the user will generally only need one representation, but the client may switch from one presentation to another depending on changes in situations and / or conditions. For example, if the bandwidth is high, the streaming client can choose a high quality, high bitrate representation. When the bandwidth is reduced, the client can adapt to these situations by switching to a low quality, low bit rate representation.

Switch points (or random access points) are samples of a presentation from which decoding of media samples can begin without requiring information of the data preceding the stream. More specifically, in the case of video representations, not all samples are random access points, since samples (frames) generally depend on the preceding frames. If the streaming client wishes to switch presentations, it must begin decoding the new representation at the switch point to avoid wasting effort. In some cases, switch points are signaled to the segment index (sidx) as a streaming client.

A representation group (often simply abbreviated as a group) is a set of switchable representations. A media presentation may include more than one representation group. For example, it may have one representation group for video representations of different bit rates and another representation group for audio bit rates. In the DASH standard, a representation group is often referred to as an adaptation set.

A segment is a file that contains media data for at least a portion of one representation. A fragment is part of a segment that is mapped from the start p-time of the fragment to the byte range of the fragment within the segment. Sometimes, the term subsegment is used instead of a fragment, and they can be considered equivalent. Some media content is not divided into fragments, in which case " fragments " may be referred to as segments themselves.

3 is a timing chart showing two possible presentation switching processes. The switch can be backward looking (the first process; FIG. 3A), where the switch point in the switch-to representation is already in the switch-from representation It is found by routing in the requested p-time stretch and selecting the previous switch point in the p-time reverse direction from the switch-to-nearest presentation closest to the end of this stretch. The second process (FIG. 3B) is forward: d finds the next switch point in the p-time forward direction in the switch-to-tell presentation starting from the last requested p-time in the switch-forward presentation.

4 is a timing chart showing the switching processes when the switch point is aligned and when the switch point immediately follows the last requested fragment. Since both processes operate identically in such a configuration, the diagram illustrates the operations of both the reverse and forward methods. Thus, when the switch points are aligned, no process can download the overlapping data.

Presentation time is the period of time during which media is expected to play at normal speed, playout, or playback. For example, a 30 minute video presentation will play for 30 minutes. The user may fast forward or rewind, which will change the actual time taken, but it should be understood that the presentation is still a 30 minute video presentation. The presentation element provides a presentation to the user over the presentation time. Examples of presentation elements include a visual display and an audio display, or a video / audio stream sent to a device capable of presenting it. "Playback" is a term used to describe the consumption of media. For example, a smartphone may download or acquire media that re-present the presentation during presentation time (p-time) of the presentation, buffer it, and the media player "consume & It is said that the buffer is not completely empty until at least the end of the presentation time so that the receiver does not experience a stall in the presentation while waiting to acquire more data. Of course, " playback " or " playout " does not imply that the media is played more than once. In many cases, once the media is consumed, it is never used again.

The presentation buffer is a memory element that can access the receiver, the media player, or either or both. For convenience of description, we use the terms "presentation buffer", "buffer", "media buffer" and "playback buffer" interchangeably and refer to data that has been downloaded but has not yet been played out or consumed, Lt; RTI ID = 0.0 > media data. ≪ / RTI > Data containing the presentation buffer can be partitioned between different components within the device, i.e., some portions of the downloaded data are held by a first process, e.g., a receiving process in the device, It may have already been delivered to another process, e. G. A playout process in the device. Also, at least a portion of the data containing the presentation buffer may be at least partially replicated across other buffers or other processes. In some cases, not all of the downloaded but not yet-played-out data is still considered to be in the presentation buffer, for example, in some cases once the media content is passed to the media play it is no longer in the presentation buffer It may be considered not. In general, the amount of media data that is downloaded but not yet played out and is not yet considered to be in the presentation buffer is very small, even if it is.

A presentation buffer accommodates unevenness is receiving and playing back media, storing received media data until it is consumed. After the media data is consumed, it may be deleted or kept stored, depending on the configuration. In some implementations, the size of the presentation buffer (which may be measured by the number of bytes of data that can be stored in the presentation buffer) may change over time. For example, the presentation buffer may be dynamically allocated as needed from the shared memory.

In many of the examples described in detail herein, the presentation buffer can be considered to be characterized by size. In the case of a fixed memory dedicated to a presentation buffer, its size may be measured by the number of bytes that can be stored in the available memory. In the case where the presentation buffer is dynamically allocated, the " size " viewed as the presentation buffer may be the number of bytes currently allocated to the presentation buffer, the maximum number of bytes that can be allocated to the presentation buffer, have. The presentation buffer size may also often be measured to the presentation time playout duration of currently available media in the presentation buffer.

The presentation buffer also has another feature, its "level" or "fill level". The level of the presentation buffer indicates how much unspent data is present in the presentation buffer, e.g., measured in byte or presentation time duration. The level is expected to increase as media data is received and to be lowered as it is consumed. The level is only logical - for example, the presentation buffer is still full of media data, but some media, such as already consumed media data, is marked for overwriting as new media data is received. Some receivers are in a state when there is uncommitted media data where the "empty buffer" is zero and a "full buffer" is programmed to be in a state when 100% of the presentation buffer is filled with uncommitted media data have. Other receivers may have different boundaries to vary over a range of 0% to less than 100% of the presentation buffer size. In the case of a presentation buffer that is uniquely assigned when shared memory is used and uncommitted media data is stored therein, the presentation buffer is, by definition, always full, It may be meaningless to use the size of the denominator as a denominator. Instead, the level of the presentation buffer may be measured as the ratio of the amount of uncommitted media data in the presentation buffer divided by the maximum size allowed for the presentation buffer.

1. Overview of client components

Referring again to Figures 1-2, various components of an exemplary client are shown.

The SC keeps track of the metadata, such as information about which representations are available and what their fragments are. The SC is also responsible for buffering media data received over the network and handing it over to the media player. The SM is responsible for making decisions about which representations should be downloaded at some point in time and for rate switching. Finally, the RA is responsible for downloading the media fragments, given exact URL and byte-range information as provided by the SC.

The SM is a software component responsible for rate switching decisions. One of SM's goals is to pick the best content for a given situation. For example, if a lot of bandwidth is possible, a high download rate may be obtained, and therefore the SM should choose a high-rate presentation. If the download rate drops significantly, the selected high representation may no longer be maintained, and therefore the SM should switch to a lower representation rate that is more appropriate for the situation. The SM must avoid being completely emptied of the playback buffer (which may cause a playback stall), but at the same time switch too fast or too fast to switch too often. Moreover, it aims to request the highest quality content that can be downloaded through the network and played back without stalling. The SM can be extended to consider factors other than download speed in the decision process. It can potentially take into account such things as battery life, display size, and other factors when making a presentation decision. Such additional constraints may be added as filters to the SM and do not affect the underlying rate decision calculations of the present disclosure.

Representative, high-level, client operations will be described below. It is assumed that the user requests specific media content, such as a live sports broadcaster, a pre-recorded movie, an audio stream, or other audio-visual or other content, which may carry video and audio and other types of media. The client will supply the request to the SM, perhaps via a user interface or computer interface. The SM will request the SC and receive indications as to which representations are available, which p-time spans are covered by which fragments, and where the switch points of the replies are located. In addition, the SM can have any available information about the short-term download rate - as described below, the RA reports this data to the SC and the SC reports or provides it to the SM.

The SM uses the information together with its past history to estimate the sustainable rate and determine the amount of media content to download from the appropriate switch point in the presentation and from the presentation starting from that switch point. As the downloads proceed and the media content is played back, the SM uses the supplied information to determine whether the rate switch is appropriate or not. If the rate switch is not appropriate, the SM informs the SC to continue fetching the fragments from the current representation. If the rate switch is successful, the SM determines the potential switch points and determines which fragments need to be fetched from certain representations in order to make the desired switch. The SM then forwards the information to the SC. Whenever a decision on the next section of video to be downloaded should be made, this exchange between the SC and the SM is performed periodically. To make a good decision, the SM monitors the buffer level and the SM can determine if the buffer is full and if fragments need not be downloaded for some period of time.

Once the SM has determined which fragments to download, the SC will cause the RA to actually download the fragments, keep the downloaded fragments in the media buffer, and finally deliver the media data in the media buffer to the media player when it is time to play it out Be responsible for.

The SM is no longer actively involved in those fragments that it has requested to download to the SC. However, even after the downloading of a predetermined fragment has already begun, the SM can change its mind and cancel the previously requested fragments request. This feature is useful if the download rate drops dramatically and the fragment being downloaded appears to be unavailable until the media buffer is completely emptied. When such a situation occurs, the SM must detect it and cancel the request and instead switch to a more appropriate rate.

Once the SC receives a fragment handle to fetch from the SM, it finds the URL and byte range of the corresponding fragment in the data structure and uses it to generate a request that it passes to the RA. It is also responsible for retrieving the response data from the RA and transforming the received media fragments into a playable stream. Finally, the SC is responsible for parsing and grasping metadata, such as playlists, in the case of data obtained from the MPD, segment index (sidx) boxes, or Apple's HTTP Live Streaming (HLS).

The RA is a component that takes fragments and metadata requests received from the SC, generates corresponding HTTP requests, transmits them over the network connection, and retrieves the corresponding responses and forwards them back to the SC. A network connection may be an Internet connection, a cellular-based connection, a WiFi connection, or other network connections that can handle HTTP requests and responses. A network connection may be embedded within a single device, i. E. It may be an internal interface to media data already cached in the device. There may be many combinations, that is, some media content may be downloaded from a wired Internet connection, some via a cellular based connection, some via a WiFi connection, and some from a local cache. In some cases, the connections through which the media data is downloaded may be mixed, some through cellular, some via WiFi, and some through wired. Certain requests may differ from HTTP in some cases, but HTTP is preferred if the servers serving the media content are HTTP servers.

In its simplest form, the RA is an HTTP client. However, it may be desirable that the RA be more efficient than a normal HTTP client. One goal of the RA is to achieve a sufficiently high download speed; It should aim to download significantly faster than the selected playback media rate. On the other hand, it should also be noted that the timeliness of raw throughput is not compromised: the fragments to be played out sooner are more urgent than the later ones, and the RA should try to receive them timely . Thus, for timeliness, you may need to sacrifice some time spent. The RA should be designed to work well in all reasonable network situations.

The basic design of the RA is to use several connections and possibly forward error correction (FEC) to get the best results. Thus, an RA will typically need to handle more than one open HTTP connection. The RA will send out requests on those connections. The RA may, in some circumstances, divide the requests into a set of smaller requests. Upon receiving the corresponding responses, the RA aggregates the data back into a coherent response. In other words, the RA is responsible for determining the granularity of HTTP requests to send, determining which connections to dispatch requests to, and which parts of the source fragments or repair segments to request. The granularity of the requests may depend on many things, such as the buffer level, the urgency of the request, the number of connections available, and so on.

Each request originated by the RA is an HTTP request to the metadata, or to part or all of the fragment request sent by the SC to the RA. It may be repair data generated from source media data or source media data. Responses to RA requests generated from an SC fragment request should, in most cases, be sufficient for the RA to reconstruct all the media data of the fragment request, and the RA can then pass it back to the SC. Thus, the RA is responsible for assembling the responses from the RA requests associated with the media fragment request in response to the fragment request provided to the SC again. For example, if there are some RA requests for FEC repair data, assembling by the RA may include FEC decoding.

In addition to handling HTTP requests, the RA measures the download speed during short-term periods during time slices of a certain sampling rate. The exemplary sampling rate is 100 ms, i.e., RA measures download rates during 100 ms periods. This data is used by the SM to calculate its download speed estimate and eventually to make rate determinations. Other sampling rates are of course possible.

The RA does not need to know about metadata or segment structures such as the DASH Media Presentation Description (MPD). In certain implementations, the RA implements an HTTP retry over several connections, even in some cases across different types of connections to similar or different servers, using several concurrent instances of the HTTP stack implementation .

The RA is responsible for knowing when the SC can accept new requests. The SC invokes the SM to determine the next fragment to request and provides the RA with the appropriate request. The RA also provides some status information. The RA can regularly provide the SM with the short-term download speed and the total time spent on the download via the SC. The SM can also indirectly poll the RA for this information via the SC. In addition, the RA also informs the SM what percentage of each individual request has already been completed. This information is provided similarly to the API that the SM calls to retrieve it.

There should be as little buffered data as possible in RA or SC (except for deliberate media buffers), and very tight data flow between RA, SC, and physical media pipelines. The same is true for various types of HTTP requests; It is necessary to make a decision on the fragments to request in only a small amount of time earlier than when the corresponding HTTP requests are transmitted over the network. One reason is that if the SM has to make further decisions about the request in advance, the information is less accurate and recent, and consequently the decision will be of lower quality.

The SM submits the requests to be issued in turn. However, the SM may issue new requests even if all previous requests are not completed; Coexistence requests are allowed. The SC sends the RAs to the RA to issue them. The RA then processes the coexistence processing and confirms that it passes the received data back to the SC.

Coexistence requests allow RAs to implement HTTP pipelining. In fact, even RAs using multiple connections are well suited to this scheme.

1.1 Stream Manager (SM)

The SM determines when to request fragments and which fragments to request in response to a combination of user actions, network conditions, and other factors. If the user decides to start watching the content, the SM is responsible for determining the first fragment to request for that content, starting from the p-time specified by the user or by the service provided. For example, some live streaming may require all users to be viewing the same p-time portion of media content at the same r-time, while other live streaming and on-demand services may require some p- You can allow end users or applications flexibility to play back-time. When the media buffer is full, the SM temporarily stops providing additional fragmentation requests. The SM is responsible for determining what quality to play the content at each point in the p-time, based on other factors such as network conditions and display size, remaining battery life, and so on.

If the SM determines that it is appropriate to provide the fragment request, the SM may provide the request only if the RA is ready to receive and process the fragment requests. The SC determines when this is true by polling the RA and forwards this information to the SM.

If the RA is ready to receive the next request, the SM determines if the new request should be an issue and selects the next fragment to request. The SM requests one fragment at a time for media data. The SM is responsible for requesting fragments to allow timely seamless playback of the content. Playback changes of representations may generally only occur at switch points, and there may be multiple fragments between two consecutive switch points; The SM considers its limitations.

In general, the SM only attempts to request reasonable fragments to believe that they will be received in a timely manner for smooth playback. However, considering that network conditions can often change very rapidly, this can not be guaranteed in all situations. Thus, the SM also has the ability to cancel requests. The SM will cancel the requests if there is a risk of serious stalling if the congestion is detected and no action is taken. For example, if the download rate suddenly drops suddenly due to deterioration of the network situation immediately after the fragment request is issued, there is a possibility of stalling if no action is taken.

The SM keeps track of the representation, R, and the p-time end, E, of the most recently selected fragment. The SM typically chooses to request the next fragment with a starting p-time of E '= E. Some changes may cause the start time to be determined from the buffer level and the current playback time.

The SM may generate a sequence of requests intended to produce a stream that can be smoothly played back if potential overlaps are discarded at switch points. The order in which the SM generates these requests is the order in which the RAs should prioritize them (though not necessarily necessarily). This is also the order in which the RA passes the received data back to the SC and the SC should play it out.

If the SM determines that it needs to switch the rate, there are usually two processes to do so. In one process, the SM finds a switch point (also often called a "random access point" or "RAP") P in a new "switch-to" Once the point is identified, the SM begins requesting fragments of the new representation. The second process is to find a switch point, P, that has a p-time that is later than or equal to that of E, and to wait until a fragment with an end-time that exceeds P is requested, Keep asking. In either case, it may be useful to signal switching to the SC.

Note that all of these processes are characterized in that some overlapping data may need to be downloaded. There is a p-time stretch that may need to download data for both switch-based presentations and switch-to-present presentations.

Which of these switching processes is appropriate depends on the situation. For example, in certain situations, the overlap is excessively large for one process, while for another it is very short. In the simplest case where all fragments are aligned across the representations and all fragments start with RAP, these switching processes are reduced in a simpler way, where the SM is switched to a switch-to- And switches by requesting the next fragment from the presentation. Note also that in this case no overlapping data need be downloaded.

1.1.1 SM Fragment Determination Process

This section describes the SM fragments decision process to determine which fragments to request from the SC. In these examples, a single representation group is assumed, but examples include video representations using, for example, a video representation group using a plurality of representation groups and audio representations Selecting a transaction can be extended to describe processes.

The next fragment selected by the SM will typically have a start p-time that is the end p-time of the previous fragment request. Some detailed logic is described below that may be implemented in the SM to select the next fragment to request.

In the following examples, fragments are assumed to start with RAPs and be aligned between representations. Unless this is true, variations of this explanation are possible. If there is such a condition, the fragment determination of the SM is reduced to a rate determination, that is, the SM determines whether to remain in the current presentation or switch to another. In the more general case where fragments may not necessarily be aligned across the representations and may not start with RAPs, the decision is similar, but the switching cost is higher, which may be considered.

The SM representation process includes two logically separate processes. The first process is a rate estimator, which is a decision process that calculates the approximate download rate maintained from the short term samples provided by the RA and the second process makes switch decisions using this estimate.

2. Rate estimation process

The adaptive bitrate streaming client typically uses a download rate estimator that is later used by the rate determination module to select the correct bit rate media. In this approach, if the download rate is high, high quality media can be streamed. The change in the download rate may trigger the presentation switches. The quality of the rate estimation has a great influence on the quality of the streaming client.

A good quality rate estimator for adaptive video streaming devices should have many characteristics. First, it should have little change even if the short-term download rate changes a lot. Second, it must adapt quickly to rate changes on the underlying channels. If the channel rate drops significantly, the estimation should quickly reflect that fact so that the device can adjust the quality accordingly without stalling. Correspondingly, the rise in video quality should be observed quickly so that higher quality content can be fetched.

Satisfying both of these conditions may require trade-offs. In general, estimators with small changes will have a large response time and vice versa. For example, consider a simple estimator that can be used in a device. The estimator will take a moving average for the last X seconds of the download for any fixed X. For example, choosing a large X of X = 30 seconds (s) will result in a relatively flat estimate with little change, but will only react slowly to the download rate changes. When such an estimator is used for rate determination, the resulting player may fail to stall frequently to a drop in bandwidth or to do so in a timely manner at a higher bit rate in a secure manner. For these reasons, an implementation may choose a smaller X, say X = 3s. Such a choice would undermine stability but result in faster rate adjustments. The rate estimate will vary greatly, and thus the player may change the video playback rate very often, resulting in a bad user experience.

In Fig. 5, the rugged curve is a raw download rate with many short-term fluctuations. The rate estimator is a flattened version of the rugged download rate. At a rate change, it converges to a new rate that lasts and remains similar as long as the rate does not change.

One of the desirable characteristics is that if there is a small buffer level, the adjustment is fast, which results in a fast adaptation of the rate so that the presentation buffer is not empty before adjustment when the download rate is falling. On the other hand, if there is a lot of media data in the media buffer, the rate estimate should be smoother with a slower adjustment. If there is more media data in the media buffer, the playout rate should tend to remain higher for a longer period of time when the download rate falls than when there is less media data in the media buffer.

The rate estimation process described below, referred to as the pker , pker process, or pker -type process, not only responds quickly to rate changes, but is also stable and satisfies all of the conditions for low variability and high reactivity.

2.1. pker  process

This section describes a rate estimation process referred to herein as a pker , a pker -type process, or simply a & quot ; pker process. &Quot; The basic rate estimator uses the estimates based solely on short-term rate measurements and calculates a longer running average therefrom using one method or the other. The aforementioned moving window average ("MWA") is an example of such a process.

6-7 illustrate the effects of using a non-adaptive (fixed coefficient) exponential weighted average for rate selection. These plots, for simplicity, the new rate estimate immediately triggers a new download selection (i.e., the fragments are relatively small), and the new rate selection is only rate estimation.

Figure 6 shows the r-time aspect. As shown here, the x-axis is the download time (real time). If a dramatic rate increase occurs at time T1, the buffer begins to grow very quickly because video data is being downloaded much faster than it is being played out. The EWMA estimation gradually converges to the room rate.

Figure 7 shows a p-time aspect of the same event. In the figure, line 702 represents the bit rate displayed on the screen. The rate is adjusted much slower than the r-time plot of FIG. The convergence rate for p-time is slowed down by the NR / OR factor at the start (because the player has received about video NR / OR seconds per second downloaded at that point) compared to r-time. Thus, the net effect is that the media can play out at a much lower rate than the download rate during a significant amount of p-time when using this type of rate estimator.

If the rate is estimated for media streaming purposes, the estimator may use other appropriate information. More specifically, the download history of the media player, which includes information about how long it took to download each media segment, or the buffer of the media player, or generally, buffered or already played out More past) are the objects of interest.

One implementation uses, for example, MWA, but can select the window size as a function of the media buffer.

If the buffer level of the media player is high, the player does not face the risk of stalling, and a long term estimate can be made using a large window, which will result in a more reliable estimation. Conversely, if the buffer level is low, the player must react quickly, suggesting that a shorter average window is a better choice in this case.

Thus, one implementation of the rate estimation process uses a varying window width, using an r-time window that is proportional to the p-time amount of the current media buffer (i.e., the current amount of p-time that has been downloaded but not yet played out) It is possible.

Other implementations may choose the window width in proportion to the number of bytes currently stored in the media buffer.

An implementation may also examine the contents of the buffer itself, not just its level. For example, if it determines that a large portion of the buffer has been downloaded in a shorter time than the playback duration of that same content, this implies that the download buffer is growing rapidly, and therefore the rate estimator needs to adjust the estimate You can decide.

Similarly, the rate estimator may track the rate change of the buffer level and quickly change the buffer level according to indications that the rate estimate needs to be adjusted quickly.

Figures 8-9 illustrate operation in the same scenario as Figures 6-7 where a variable window size weighted moving average filter is used. In the examples, the " pker " process is described in programming code such as a variable window size WMA filter. The pker process may be implemented with program instructions that are executed by the processor.

In FIG. 8, line 802 is a pker rate estimate when the base channel has a sharp rate increase from rate OR (old rate) to rate NR (new rate). The amount of r-time required to select the rate to adjust at the new rate is proportional to OR / NR. The larger the increase, the faster the adjustment takes place in real time. As shown, at time T2, Buff @ T2 = 2 * Buff @ T1 and T _fast = OR / NR * Buff @ T1.

Fig. 9 shows a playback operation in p-time. The pker estimator takes approximately one buffer duration (the amount of p-time that was in the buffer when the rate increase occurred) to adjust to the new rate, i. e., the media buffer has a p- The pker estimator adjusted to the new rate until the time that it has the amount of content, where B is the p-time duration of the media content of the media buffer at the rate increase to the new rate.

A specific process for doing this will now be described. The process discards how much r-time it took to download the last γ _T - portion of the playback buffer, where γ _T is the constant chosen constant. For example, it may be the complete time taken to download the entire current playback buffer ( ? _T = 1), or the time it took to download the last half of the playback buffer ( ? _T = 0.5). γ _T > 1 is also possible. Let T _{fast be} the amount of r-time taken to download the last γ _T - portion of the playback buffer. The estimated download rate can be calculated by estimating the download rate for the previous T _fast seconds prior to the download time. Note that a different value of γ _T is possible. Other values, as described herein, contribute to other goals.

T _fast This kind of windowed average across width windows has a notable characteristic that quickly detects the rate increase. In fact, if a value of y _{T &} lt; 1 is used to determine T _fast , then the estimator estimates that if the rate increases by an arbitrary factor at any moment in time when the p-time duration of the media content of the media buffer is B, Lt; RTI ID = 0.0 > at most < / RTI > times B before converging at an increased rate.

More sophisticated rate estimation methods can combine the two approaches described above. It can in particular use the minimum values of the buffer levels B and T _fast as the average window width, i.e. the amount of r-time averaging the download rate. More generally, the download rate can be averaged over the previous r-time of the minimum of [ gamma] _B and T _fast , where [ gamma] _B is a suitably chosen constant. Such a choice would have the property that it will respond quickly when there is a risk of stalling a rate drop, and in such a case, B is the minimum and the average is r-time proportional to the p-time duration of the media content in the media buffer And thus the rate estimate will be the new rate until the media buffer is half-weighted. For example, when the rate decreases, the download rate decreases so that the media content duration of the media buffer is B and the download rate is a < 1 of the playback rate of the selected presentation before the download rate decreases, Assume that the playback rate of the selected presentation does not decrease until the rate estimate decreases to the new download rate. Then the buffer level is B '= B - x + alpha x, i. E., X p-time is emptied from the media buffer and alpha x is equal to It is downloaded to the media buffer. At a point in time where x = B ', i.e., at a point in time at which the media buffer level of p-time equals r-time at which the download is at the new rate, the rate estimate will be the new rate, Since at this point in time the estimate will be the new rate for the previous r-time of the download. Solving for x the equation x = B '= B - x + x x yields x = B' = B / (2-a), i.e. the rate estimation is performed when buffer B 'is still at least B / 2 You will reach the new rate. Instead, if the rate significantly increases at some point in time, T _fast will be the minimum and the average download rate over the previous T _fast r-time will be significantly higher than the average over the previous B r-time.

We now elaborate an example of the pker rate estimation process based on this configuration. It uses short-term rate measurements, which can be obtained from a download module such as a request accelerator (RA), and buffer information for calculating an estimate. The buffer information is used to determine the window width of short term rate measurements to obtain useful estimates.

Figure 10 shows how the pker rate estimator draws a conclusion when the download rate drops sharply. As soon as the rate drops, the buffer level begins to drop. The rate estimate also begins to adjust. At the latest, the rate estimate reaches the new rate NR until the buffer level drops by a factor of two. For example, there is no rate determinations in the middle and Buff drops linearly. If there is an intermediate decision, the drop of Buff will be gradually slowed down.

The design goal of the pker process is to use an average window large enough so that it does not have noise numbers, but numbers that are short enough for it to respond. The pker process achieves this goal by using a windowed average with a window size that varies dynamically. RA is the rate at which the level of the playback buffer (in p-time), the process parameters γ _B and γ _T , T _fast , and the r-time taken to download the last γ _T - for use by, pker process including the average download rate of the last C the duration of the download from the stored value, r, r-time holding a few variables to the memory, wherein, C = max (STP, min (γ _B · B, T _fast ,)) and STP is the smallest acceptable window size, which must exceed the sample time period (eg, 100 ms). In some embodiments, γ _B = 1 and γ _T = 0.5, but other values are possible, resulting in a qualitatively similar behavior, as long as both are positive and γ _T > 1. A small [ gamma] _B causes the pker process to respond quickly to rate reductions, while a small [ gamma] _T causes the rate to quickly respond.

As described herein, to calculate the download rate during the C duration, the SM uses the download rate information periodically provided by the RA. For that purpose, the SM can keep a history of the download speed information provided by the RA. The duration at which the average is taken is at most the gamma _B buffer duration and effectively limits how much history needs to be retained when there is an upper bound on the media buffer level.

If the amount of time it takes to download a stream same as that taken for it played out, and we therefore have a T _fast = γ _T · B, if the selected play-out rate equal to the approximate download rate, buffer values, C is approximately buffer Note that this is a duration. Since choosing roughly the buffer level amount at r-time is a natural choice for the smoothing interval for download rate estimation and if it is desired to avoid stalling, the amount of foresight the streaming client should have to be.

In a simple implementation, the average window width is proportional to B, the amount of p-time contained in the video buffer. If the download rate is a multiple of the rate of the selected media, then every second of the downloads results in k of p-time of the media being downloaded, which means that the rate estimate is adjusted very slowly . For example, if k = 10 and there is a buffer of 10 seconds, the rate estimator will download p-time of about k * 10s = 100s before adjustment, which is a very long time. This gives motivation to introduce T _fast in pker methods. In fact, if the exponentially weighted moving average is used for flattening, things can even get a little worse because those filters have an infinite impulse response. For this reason, the pker process uses a finite impulse response instead. The normal moving average works effectively; Implementations may also use more sophisticated weighted moving averages.

Figure 13 shows this last point. It shows a comparison of a single (fixed-width) moving window average and an exponentially weighted moving average. When the graph shows that the rate change is visible, the fixed window moving average will initially converge slower at the new rate, but it will converge within one window duration. The exponentially weighted moving average tends to move quickly at the beginning, but converges only slowly at later stages. Unlike a windowed moving average, it does not converge within a fixed window, but instead takes an algebraic time to the magnitude of the rate change to converge.

For γ _B = 1 and γ _T = 0.5, the pker process provides various assurances. First, if the download speed drops by any factor, the estimate is adjusted to the new download speed within the time it takes for the buffer to reduce to half its original duration. On the other hand, if the download rate is increased by an arbitrary factor, at most one additional p-time of the buffer value will be downloaded before the pker process converges to the new rate. Smooth calculations will show that similar constant-fraction guarantees are maintained for any choice of 0 < ? _B and 0 < ? _T <1.

One approach for calculating the buffer level, B, is as follows. Let T be the current playback p-time of the media player, F _{i, 1} , ... , F _{i, n} are fragments that have been downloaded or are sorted out in ascending order of start time and have not yet been played out in the representation group i. Any fragments of group i that are still being downloaded are F _{i, 1} , ... , F _{i, n} . Let the ratio of α (F _{i, j)} a fragment of bytes F _{i, j,} divided by the size of the already downloaded fragment F _i, such as the number of bytes of _j, the downloaded fragment F _{i, j.} The values for the various i and j can be calculated by the RA and passed to the SM. For a given group i , we define Equation 1 as the current total amount of downloaded p-time.

(식 1)

(Equation 1)

To calculate the overall T _P -value from the results of Equation 1, the DASH client calculates the weighting factors w 'of each group, which are determined from the number of media presentation description metadata (MPD) and representation groups, G And performs the calculation of Equation (2). The buffer level B is then defined as B: = T _p -T .

(식 2)

(Equation 2)

Equation 2 also captures the portion of the buffer that belongs to the fragments that are currently being played out. This definition also works well when several fragments are concurrently downloaded.

의 히스토리, H를 보관하고, 여기서 K-번째 관측이 마지막이다. 우리는 히스토리가 관측 순서로 저장된다고 가정하고; 따라서 우리는

와

및

또한 갖는다.To calculate T _fast , the SM maintains some history in the normal case. Let T _{r be} the total amount of r-time that RA spent trying to download (try), and let Z be the total amount of bytes downloaded by RA. The value of Tr is calculated by RA. SM is i = 1, 2, ... , Sampled at regular intervals (e.g., every 100 ms) for K ,

, H, where K-th observation is the last. We assume that the history is stored in observation order; Therefore,

Wow

And

.

Now, to calculate T _fast , assume that B has already been computed in the manner described above. Next, the RA determines j to cause the inequality of Equation 3 to be satisfied, for example, by searching the history with a binary search.

(식 3)

(Equation 3)

이다. 무한 히스토리를 보관할 필요는 없고, T_i 값들이 최대 버퍼 지속시간의 γ_B 보다 많이 스팬(span)하는 것으로 충분함을 주의해야 한다.Next

to be. It is not necessary to keep an infinite history, and it should be noted that T _i values span more than γ _B of the maximum buffer duration.

Along with the enlarged modification of FIG. 16, FIG. 15 shows how the B and T _fast values used by the pker process can be determined from the history of the recorded (T _p , T _r ) values. The drawing shows that r-time and p-time are equally fast (there is no interruption to downloading), so the playback time (p-time) is the 45-degree slope of the download time (r-time). (T _p , T _r ) - The history of values can be plotted as a curve directly above the playback time line, if no playback stall has occurred in the graph. Buffer level B is then the difference in the last recorded T _P - value for the playout time. The value of T _fast can be represented in this graph by measuring the horizontal distance from (T _p , T _r ) - curve at the level of γ · B below the current (last) T _P - value.

Figure 11 uses a presentation of the kind shown in Figures 15-16 to illustrate the responses of the pker process to sudden rate increases. T _fast is relatively small when the reception rate experiences a sudden increase that the player has not yet reacted to. It exhibits fast response at a high receive rate. Note that the entire average window is within the high rate portion of the graph, as it is relatively narrow. Thus, at this point, the pker estimate has already converged to a longer rate.

Figure 12 again uses the presentation of Figure 15 to illustrate a variable window size WMA filter (e.g., pker ) response to a rate drop. In this case, T _fast becomes relatively large, but the buffer is emptied, and therefore B becomes small, allowing the entire average window to fall within the low-rate region after some period of emptying. As shown, the averaging window width, B is B is smaller than T _fast, estimated before the buffer is completely non will still converge to the new lower rate.

14 is a flowchart of a pker rate estimation process.

Once the T _fast and B values are calculated, the C value is readily available and the final step is to calculate the rate R over the last window of duration C. [ For that purpose, the Z ⁱ and T ⁱ values of the history are used.

인 가장 큰 j를 찾고, 다음에 (2) 식 4에서와 같이 평균 다운로드 레이트를 계산한다. 제 1 단계에서 그러한 j가 존재하지 않으면, SM 또는 RA는 j:=0, 즉 가장 오래된 공지의 관측으로 설정한다. j 값은 이전 검색에 의해 효과적으로 결정될 수 있다.To calculate the rate over interval C, SM or RA does the following: (1)

(2) Calculate the average download rate as shown in Equation (4). If such j does not exist in the first step, then SM or RA is set to j: = 0, the oldest known observation. The j value can be effectively determined by the previous search.

(식 4)

(Equation 4)

Each group has an associated weight, w, corresponding to the ratio of the total bandwidth that the group is expected to consume. It is preferably a function of the information provided by the MPD after the useless representations have been removed. Here, the proposed definition of the weight w of group g is w (g): = maxrate (g) + minrate (g), where maxrate () is the maximum playback rate in group g and minrate It is.

From the weights w, the SM or RA may calculate the normalized weights w 'as follows: Client is group 1, ... , G, the normalized weights are the weights divided by the sum of all weights, as in Equation 5. &lt; RTI ID = 0.0 &gt;

(식 5)

(Equation 5)

It is contemplated that the normalization is performed on the weights actually being streamed. For example, if you have a group that is not being streamed, it should not count.

There are some assumptions in the operation of the pker process. For example, the buffer levels of the different representation groups should be kept relatively close to one another. The pker process works better that way. For example, suppose a group has a very large buffer, the other has a very small buffer, and both have similar weights. In such a case, it may be necessary to adjust the rate estimate quickly, since in the case of small buffers it is necessary to avoid stalling at the time of the situation change. However, the pker process will still flatten the estimation as it does for a much larger buffer. Conversely, in the case of large buffers, the measurements will have somewhat higher fluctuations, which the buffer level allows, and therefore will result in sensitive rate decisions.

In some cases, it is inevitable that the representation groups have a large difference in buffer level. For this reason, other implementations may use a modified pker method that adjusts rates faster when certain buffers are very small, and thus can better protect bits against stalls in such cases. Such an implementation is calculated in the same way as before the T _fast, but the window size = max C (STP, min _(fast T, _{p T,} -T _1, T _p, -T _2, ..., T _{p, N} -T )).

Another variation of these download rates estimates involves making decisions for that group using a pker estimate that is independent for each representation group.

3. Fetching Strategy

Streaming video players also typically have limited media perfor- mers. Thus, in normal operation, the buffer full state is expected to arrive some day. When the buffer reaches a full state, the streaming module must block the media input to avoid overfilling the buffer. An easy way to do this is to wait until the buffer is empty enough to hold the next fragment, and then start fetching again whenever the buffer is full.

The effect of this method is that each fragment will be fetched individually and there is a time interval between each fragment request, that is, the amount of time it takes to empty enough buffers so that the next fragment is eligible and requested.

The TCP protocol automatically adjusts its download rate based on the current network conditions. When a download is initiated via a TCP connection, the initial download rate is generally very slow and increases as the TCP protocol checks to see if a higher download rate can be achieved. Whether TCP schedules download rates, and whether TCP generally responds to the characteristics of TCP-to-TCP connections, is a very complex and unique end-to-end network latencies, buffers of network elements that follow TCP serving and acknowledgment paths Capacity, contention traffic along these paths, what variants of TCP are used, and so on. In general, TCP starts with a slow download rate and increases its download rate over time, so that the average download rate of TCP connections over the entire download time is only approachable at a sustainable TCP download rate do. For example, if the sustainable TCP download rate is 1 megabit / second and the TCP connection starts at a download rate that is essentially zero and increases linearly over time to 1 megabit / second over one second, The average download rate is 500 kilobits / second, and the average download rate takes 10 seconds to download to achieve 95% of the download rate that is sustainable. For this reason, a fetching strategy with many download gaps is not ideal, where download gaps are time periods between the completion of one download request and the start of the next download request. Even when the gap between download requests is zero, it is generally non-ideal because TCP typically takes some time to complete the previous request and increase the download rate for the next request. After each gap, a sustainable throughput may have to be achieved again, which reduces the overall achieved average download rate.

Such a reduced rate can lead to a smaller rate estimate, and thus a smaller media rate. This in turn allows smaller (in byte size) media fragments to be downloaded, which further increases the relative size of the gaps and potentially a much smaller playback rate. In other words, the effect self-amplifies.

Therefore, it is advantageous for the DASH client implementation to use a process that minimizes the impact of this issue.

An implementation may continue to download media data and then periodically empty the buffer level as follows: Whenever the amount of p-time requested but not yet played out exceeds a preset high watermark, M _h , the watermark M _l The SM no longer issues any requests until it falls down. In a more specific implementation, M _h = 20 seconds and M _l = 10 seconds, but in other implementations such values may be lower or higher. After a drop below the low watermark, normal operation resumes and the SM begins to resume fragmentation decisions.

Other implementations may use watermarks specified in bytes rather than presentation time to achieve a similar effect.

The fact that the buffers are periodically emptied can be advantageously used by them by other parts of the system. For example, as described in section 6.1.2, it can be used to obtain fresh estimates of the RTT.

Fig. 17 shows the operation of the &quot; watermark &quot; fetching process. The top graph is a buffer level graph showing alternating patterns of emptying periods and fetching cycles. The download rate is shown in the lower graph. At the beginning of each fetching period, TCP takes some time to reach a sustainable maximum rate, and thus the average download rate (during the fetching period) is less than the maximum achievable download rate. The greater the difference between the low watermark and the high watermark, the longer the fetching periods and the higher the average rate.

4. Rate selection process

When initiating the request for media data, the streaming module (SM) uses several methods to select the first playout rate. It may take the lowest available rate and may, for example, determine a presumption of keeping a history of network conditions and then selecting which playout rate may be maintained without stalls based on this history. If the SM is already receiving data and therefore has a rate estimate R (e.g., such as one of the rate estimates computed in the methods from Section 2) that it can use, To be changed.

A simple rate determination process will now be described. The receiver determines the highest bandwidth representation with a playback rate lower than the estimated download rate R and selects it as a representation to play back the data. Although easy, this approach has a number of problems. First, it naturally does not let the small media buffer stretch, thus doubling the stall even when the download rate changes only slightly. Second, the varying estimate R will result in rapidly changing rate decisions, which may or may not be necessary and may be visually unstable. Third, it at least approximates the start time, which is the duration of the fragment, and thus generally a few seconds.

Hence, the DASH client is able to determine its rate decisions by not only the download estimate R but also the buffer level B (i.e., the amount of p-time that has been buffered but not yet played out) and the estimate of the p- Lt; RTI ID = 0.0 &gt; D , &lt; / RTI &gt;

Thus, an implementation may select the largest playback rate that is proportional to R as the rate of decision, where the proportional factor is a function of the buffer level.

In general, the proportional factor l is an increasing function of the buffer level. For example, an implementation may make λ an affine function at the buffer level.

If lambda is a function of the buffer level, then one implementation can choose lambda to be small when the buffer is empty or small. Such a choice is beneficial because it allows small buffers to grow and also provides some safety for stalling when the download rate is not precisely predicted.

For larger buffer levels, an implementation may choose to set the value of lambda close to, equal to, or even exceed. It will ensure that a high playout rate is chosen to be downloaded when there is no immediate risk of stalling, leading to high quality media streaming in a stable state.

The rate determination process can implement a piecewise affine function of B rather than a simple affine function. The segmental affine functions can approximate any continuous functions to any desired degree of accuracy, which makes them an appropriate choice. Any other parameterizable types of functions having the same properties may be selected instead.

Other implementations can make λ a function of buffer level in bytes rather than p-time units of buffer level.

Another implementation may be the λ, as well as the buffer level B, as a function of both the frequency of the buffer level B and the switch chance. The reason for doing this is that players with fewer opportunities to change the rate will be more likely to be in the future for each decision than those who have more frequent opportunities to change. Thus, in the former case, each crystal is associated with a larger time span, and also a higher risk. This implies that in order to keep the risk of stalling low, it is better to select a lower rate in the former case than the latter when the buffer level B and the estimated download rate R are equal.

A specific method of rate selection process that takes into account the frequency of rate switch opportunities is as follows. Let D be the general amount of p-time between two consecutive switch points in the stream. The value of D depends on the encoded video and may, for example, be the maximum distance in p-time between two successive switch points, or the average distance of two successive switch points, or the 90th percentile of two consecutive switch points Number, or any other suitable measure of the p-time distance of two consecutive switch points of the media. Considering such D , the method includes choosing lambda as a B / D delineated affine function or a variant thereof, such as B / max (u, D) or B / (D + u) , Where the value u is added to account for the overhead incurred when issuing the requests. The value of u may be a small constant amount of time (such as, for example, 100 ms). More precisely, an implementation may make u a small multiple of the estimated RTT.

As with the above-described method, the process of simply determining the rate decision based on ? R has the disadvantage that even relatively small variations in R result in many rate switches. This may not be desirable. When there is enough buffer, it may be better not to react immediately to a small change in R , but instead to vary along the buffer level.

To achieve such an operation, the process can use the values of lambda and mu , both functions of the same amount (e.g., B, B / D, B / max (100 ms, D) This is for selecting a new rate decision together with the current rate. The functions should be chosen such that [ Lambda] .R is the low acceptable rate selection and [ mu] R is the high acceptable rate selection. The process can then be designed to use those two values as a guide for good rate determination.

In such a setting, the functions, generally λ ≤ μ, must be chosen.

The rate determination process may decide to keep the rate the same if the previous selection was already in the range from lambda. R to mu R. If the previous selection is smaller than λ · R, λ · R and equal to or less than The largest available play rate is selected. The previously selected is greater than μ · R equal to μ · R or smaller largest available playback rate is selected.

An implementation may choose to hardcoded the functions lambda and mu . Alternatively, it can select functions in a more sophisticated manner depending on the environment. In particular, one embodiment is suitable λ and μ functions The client can choose as a function of the amount of buffering to do as much as possible. For on-demand content, the client can choose to prefetch a lot of data, potentially a few minutes of media data. For low latency live content, the client can buffer only the amount of media specified by the end-to-end latency, which is only a matter of seconds. For content with small buffering, the client can decide to select lambda and mu functions with a more conservative, i.e. smaller value.

A specific implementation may, for example, linearly interpolate the function between two extremal functions λ ₁ and λ ₂ , where the selected interpolation point is the low buffer watermark M ₁ (see section 3). So it has two hard-coded functions, lambda ₁ and lambda ₂ , for certain values m ₁ , m ₂ , l ₁ is used for small values M _l less than m ₁ , and lambda ₂ is M _l ≥ m ₂ , where m _{1 &} lt; m ₂ . m ₁ In for the values in the range up to ₂ m, the function _{λ (x): = λ 1} (x) (m 2 -M l) / (m 2 -m 1) + λ 2 (x) (M l -m ₁ ) / (m ₂ -m ₁ ) is used.

We now describe a specific example of a rate determination process that follows the above description. To this end, we introduce some notations.

1) S ₁ , S ₂ , ... , Let S _{L be} the stream rates of the L-available representations of the presentation group (given in ascending order).

2) Let λ (x) be a piece-wise linear function that takes a non-negative scalar as input and returns a non-negative real scale day coefficient. The function λ (x) must be set at compile time or through a configuration file. For a large x , for example, for x larger than M _l , λ (x) should be unchanged.

There is an example of how such a function can be implemented. The corner points (0 ,? ₀ ), ( x ₁ ,? ₁ ), ... , ( x _N , λ _N ), where x _i is the ascending order. To compute λ (x) , find the largest i such that x _i ≤ x . Next, using Equation 6, the receiver can calculate the function.

(식 6)

(Equation 6)

A suitable example for such a λ (x) function is one defined by the exemplary parameters N = 1, [(0,0.5), (3,1)], ie x = 0 to 0.5 and x equal to 3 , And at that point the function is equal to 1 and then remains at 1.

3) Let μ ( x ) be another such piecewise linear function. An example of such a function is to compute 0 from x = 0 and reach 15 at x = 3, and keep it constant thereafter.

4) Let D be the estimate of the duration at p-time from one switch point to the next (as specified previously).

5) x: = min {( T d - T), M l} / max {D, 1 second) that, T is the current playback p-time, T _d is p-time is a rate determination, and D is As described above, M _l is a mark with a low buffer level (see section 3).

6) Let CURR be the currently selected representation (ie, used in the last fragment request). UP is at most the playout rate of the highest bit rate representation having a rate of lambda (x) R , and if there is no such representation, UP is the playout rate of the lowest bit rate representation. DOWN is at most the playout rate of the highest bit rate representation of the rate of μ (x) R , and if there is no such representation, DOWN is the playout rate of the lowest bit rate representation. In general, λ (x) ≤ μ (x), so DOWN ≥ UP in general.

Next, the rate determination process selects the rate NEXT of the next fragment as follows: (1) if UP < CURR, then NEXT : = min ( DOWN, CURR ); (2) Or NEXT: = UP.

In step 5 above, the reason for using max { D, 1 second} instead of D simply is due to RTT; 1 serves as the upper limit of the RTT.

The functions lambda (x) and mu (x) are preferably increased as a function of x . For small x , the lambda and mu functions are preferably < 1, which ensures that the selected playout rate is less than R resulting in buffer growth for small buffer levels. The selected playback rate at most equal to the max (λ (B / max { D, 1}), μ (B / max {D, 1})) · R, λ (B / max {D, 1}) and μ ( B / max { D , 1}) guarantees buffer growth for all buffer levels B less than one.

A similar process can directly select the new representation as being the best representation that the playback rate is less than λ ( B ) · R. This will still have the property that the buffer tends to fill up when the buffer is nearly empty. But it also triggers a lot of re- presentation switches, because R can oscillate very much. The more complex rate selection process described herein attempts to avoid the switches and instead allows the buffers to scale to some degree before switching down to a lower playback rate. For this to work, the functions mu and l should be chosen such that mu exceeds lambda to fit the large buffer level: if the selected playback rate is CURR and the measured rate is R , And does not allow the receive rate to oscillate somewhat without the rate switches.

(식 7)

(Equation 7)

In some versions, lambda and mu would be a function of buffer level B instead of just the ratio of B / max { D , 1}. The motivation for adopting the latter is as follows.

Let α denote the ratio of the playback rate of the selected representation to the download rate. We want to determine a good α . It takes approximately α · D r-time to download to the next switch point. Immediately before the received data is added to the buffer, the buffer will be emptied to B - α · D. To avoid stalling, we want the quantity to be positive; As a safety cushion it must also be proportional to the playback duration D of the fragment that is added to the buffer once it is downloaded, so it should be at least β · D for any β> 0. In summary, we want B - αD ≥ β · D.

Solving for α gives B / D β- ≥ α . This selection process should select a playback rate versus a download rate that does not exceed B / D - β. The functions lambda (x) and mu (x) are limits on acceptable ratios; Therefore, they must be a function of x = B / D that does not exceed x -β.

To actually account for the extra cost of the RTT to send a fragment, we replace B / D with B / max { D , 1}. More generally, 1 may be replaced by any number of estimates of the RTT, or other parameters that take into account the response time of the processes that initiate the downloading of media data from the server.

Figure 18 shows examples of lambda and mu functions that may be used to select the playback rate. The x-axis is the buffer level in units of D, and the y-axis is the playback ratio, divided by the current receive or download rate. As shown by line 1802, if the receive ratio is less than 1, the buffer will grow, and if it is greater than 1, it will decrease. 3 regions are identified. First, if the player is below the lambda -curve 1804 at the decision point, it will switch up in rate. If it is between the lambda -curve 1804 and the mu -curve 1806, it will stay at the selected rate. If it is on the μ -curve (1806), it will switch down.

Figure 19 shows an exemplary selection of ( λ ,, μ ) -functions using a "conservative" setting. This configuration is "conservative" in that it will not use all available bandwidth, but instead will stall very rarely.

Figure 20 shows an exemplary selection of ( ? ,,? ) - functions using a &quot; moderate &quot; setting. This configuration uses more bandwidth than is conservative, but it is "moderate" in that it is a little more easy to stall.

Figure 21 shows an exemplary selection of ( λ ,, μ ) -functions using an "aggressive" setting. This setting is "aggressive" in that it tries to use all available bandwidth aggressively. It will stall more frequently than the other two example settings presented.

Figure 22 shows an exemplary selection of ( λ, μ ) -functions using a process that emulates a process similar to that proposed by the MLB process, ie, some researchers working with Major League Baseball (MLB) . ( ?,? ) - functions do not change based on the media buffer fullness.

Figure 23 shows an example of side-by-side values for lambda and mu settings.

Fig. 24 shows an example of side-by-side values for lambda and mu settings.

Figure 36 includes tables of values that can be used for lambda and mu in rate selection.

25 shows a process for rate estimation, then rate-based rate selection, and then buffer management-based rate selection. In this exemplary process, one or more of the approaches described herein are used to perform rate estimation. Based on that estimate, a new playback rate may be selected and adjusted based on buffer management rules.

5. Cancel the request

In some cases, even a good rate selection process alone can not prevent video playback stalls. For example, if the download rate drops rapidly before the request is created but before it is completed, the selected bit rate may be too large and the slow download rate may lead to a playback stall before the next opportunity to change the playback rate.

In another example, the media buffer may hit the relatively low playback rate media when the available bandwidth increases dramatically due to, for example, a transition from a cellular connection to a WiFi connection. In this case, it may be advantageous to discard a large portion of the media that has already been downloaded but not yet played out, and to download the discarded p-time portions again, but this time with a higher playback rate presentation to download. Thus, the already downloaded lower playback rate media is canceled and the higher playback rate media from the other presentation is downloaded to where it will be played out, thus leading to a higher quality user experience.

For this reason, a streaming module implementation may implement a module to monitor the download rate and in some circumstances may cancel previous decisions. If the request is canceled, the streaming module must issue a new request based on an estimate of a newer, more appropriate download rate. We call this monitoring module here a request cancellation process.

The request cancellation process can cancel requests for other reasons. For example, it can cancel requests because the download rate has fallen sharply and there is a risk of stalls due to data not being received fast enough for playback. Another reason for cancellation is whether a higher quality media is determined to be selected and retrieved in a timely manner for playback. Another reason for cancellation is to determine that the stall will occur regardless of what the receiver is doing and to estimate if the cancellation will shorten the stall duration, as opposed to allowing the completion of the unfinished request. The receiver chooses the action with the estimated shorter stall and also potentially takes into account the quality of the media representation to be played back. Of course, if there is a stall and there is a stall, its duration may be different from the estimate.

The actual cancellation can be accomplished by closing the TCP connection on which the request was issued, once it has been determined. Closing will have the effect of telling the server not to continue sending data for the canceled fragment so that the bandwidth used by the closed connection can be used to fetch the replacement data.

The streaming module can then issue a request to replace the canceled fragment. For this purpose it may be necessary to open a new TCP connection.

One implementation has several options for selecting alternate requests. Which one is best suited depends on the type of content being played out.

It will try to select an alternate request that allows seamless playback of the stream. In the general case this means that the alternate request should have a switch point at or before the end time of the previously downloaded fragment.

In that case, the stall is foreseen when continuing the download without cancellation, and if the stall is avoided or at least reduced in duration, with the choice of cancellation and replacement segments, the player must cancel. It can then select the highest quality media request that has that characteristic for alternate requests.

The rate cancellation process can predict the stalls as follows: It is possible to calculate the estimated completion time of the requested request by dividing the number of outstanding bytes of the fragment by the estimate of the download rate. If the time is later than the deadline required for smooth playback, the stall is foreseen.

When an impending stall is foreseen, the request cancellation process determines whether the switch at the rate will improve the situation; Only when there seems to be improvement is the cancellation decision.

One implementation may estimate the time it takes to load the replacement fragment based only on the rate estimate and the size of the candidate replacement fragment.

Other implementations may also consider additional overhead due to cancellation: it may add a multiple of the estimated RTT to cancel the existing request and calculate the time needed to issue the new request. Data that has been queued for transmission over the network from the canceled request, but has not yet arrived at the destination may contribute to the additional delay. The client can estimate this delay by dividing the TCP receive window size by the estimated rate. Other estimates of delay may be based on the estimated bandwidth-delay product. The client can take a combination of the two estimates, as at most of the two.

In summary, the client computes the sum of the amounts required to download the entire replacement fragment, in addition to the estimate of the queuing delay, which is generally proportional to the RTT. If the stall is anticipated and the time is less than the estimated remaining time to download the current fragment, the cancellation is an issue.

The request cancellation process can also be canceled at the start if the player recognizes that it takes longer than desired to download the first fragment, since the initial rate selection was not correct.

Other rate cancellation implementations also do not allow seamless playback, but instead can select a replacement request that involves skipping a number of frames. This may be necessary in reproducing live content that requires that the delay between both ends be kept small.

Implementations that perform cancellation with a frame skip can select a replacement fragment so that frame skipping is as small as possible.

The implementation may, as an alternate request, select a top quality request that can be downloaded sustainably without exceeding a particular stall duration or skip frame duration.

Another kind of cancellation can be implemented for already downloaded fragments: if the player has already buffered some media to be played out, it will discard the previously buffered low-quality version and, at the same time, You can do so to fetch the presentation and stream it.

The cancellation process can decide to perform these substitute operations if it determines that it can receive the better quality video in a timely manner and that it can be played out without stalling

FIG. 26 shows a strong drop at a download rate occurring immediately after a new fragment request at time T1. By the time of the request, the reception rate was OR, and then it dropped to NR. The buffer level then falls. The newly requested fragment will be completely downloaded at approximately T2 = T 1 + OR / NR * fragments duration. If OR / NR is large, this is greater than the p-time duration of the media content of the T1 time buffer, which means that the requested fragment can not be played back without stalling. The pker estimator will converge much faster at rate NR but since the request is before T1, the fragment is downloaded before the estimate has the opportunity to converge to the new rate NR. In order to avoid the stall and issue a new request with a modified estimate, it is necessary to cancel the request and re-issue the request with a more appropriate bit rate.

Fig. 27 shows a case with request canceling. After a sudden drop in the download rate (line 2702), the buffer begins to empt and the estimated download rate (e.g., pker process) begins to converge to the new download rate. At some point, the stream manager recognizes that the fragment will not be timely received for stall-free playback. The point is indicated by &quot; cancellation point &quot; 2704 in the diagram of Fig. At that point, the partially received fragment will be canceled and it is kicked out of the buffer (due to further drop at the buffer level). However, after that, fragments of the correct rate can be requested, and therefore the buffer level does not drop further. In fact, if a nontrivial rate-selection process is used, it will increase again.

28 is a flowchart showing an exemplary request canceling process.

Fig. 29 shows a process for request cancel detection.

We now describe a request cancellation implementation based on the above detailed description.

In this section, N _i represents the number of fragments of the presentation group i that have been requested but have not yet been completely received. They are F _{i, 1} , ... , F _{i, Ni} . F _{i, j} is arranged in a start -p-time _high, α (F _{i, j)} is further assumed that the amount of divided by the size in bytes for the requested fragment F _{i, j} already been downloaded byte do it. The variable T represents the current playback p-time. The request cancellation detection process can proceed as shown by the pseudocode of FIG.

When the request cancellation detection process is running, it will return 0 (nil) if no action needs to be taken, or it will identify the fragments of the group to cancel. If such a fragment is identified, it means that this fragment and all of the same group following it (in p-time order) will be canceled and exported from the buffer. The SM shall then re-invoke the rate determination process and issue new alternate requests for that section.

To illustrate the process, assume that in the meantime, only one request is still outstanding. In that case, let R be an accurate estimate of the download rate, and let d _avail be the number of bytes that can still be received until the problem fragment is played out. The amount d _need is the number of bytes still missing from the fragment. Thus, if d _avail < d _need , we foresee that the player will be stalled before playing the fragment F _{i, j} . This explains the first "if" condition in the above process.

Even if the stall is foreseen, it is reasonable to cancel only if the cancellation can avoid the stall or at least reduce its duration. After cancellation, a new fragment will be selected and downloaded from scratch. If there is only one representation group and the rate determination process selects the correct rate, it will take a time of approximately l times times the duration ( Fi _{, j} ), where lambda is the currently associated lambda factor . On the other hand, if the SM decides not to switch, it will take d _need R ^-1 to complete the current fragment download. For the sake of simplicity, assuming λ = 1, we have the second condition, possibly with other factors.

6. Request Accelerator

An easy way for a streaming media client is to fetch media with a single HTTP connection. Such a client will process the fragment requests sequentially. Such an approach has some disadvantages in video streaming. First, common networking software is often tuned only for maximum throughput over long downloads. This is good for receiving many files, but it conflicts with other important streaming goals, such as a stable receiving rate. Second, due to the nature of TCP, the total capacity of the link can not necessarily be used for such HTTP connections. If the channel experiences some delay and packet loss, TCP limits the actual throughput that can be achieved, which potentially hampers the streaming client from streaming good quality media.

To avoid these problems, a special HTTP client can be implemented, which we call a request accelerator (RA) here. The request accelerator has special processes that can avoid or reduce the above-mentioned problems. The implementation of the request accelerator can achieve its goal using several key components. It can receive data using several TCP connections. Such connections may be activated in parallel. It can split data requests into smaller chunk requests, which can be individually downloaded over different connections and reassembled into one larger piece in the request accelerator. It can tune TCP connection parameters (such as TCP receive window size in particular), so that connections have an appropriate and relatively stable data reception for each other. It can dynamically adjust the number of TCP connections to use based on measured network conditions and target playback rates.

The ideal number of TCP connections to use depends on the network topology, and in particular the round trip time (RTT) and packet loss operation. Thus, RA can use methods to estimate these quantities.

The RA can estimate the RTT by sampling the time it takes from issuing the HTTP request to the beginning of the response. An implementation may use an estimate of the RTT obtained by taking a fixed period of time, say, a minimum of all such samples obtained over the last few seconds. Other implementations may use a minimum of the last obtained N samples as an estimate, where N is any integer.

Obtaining measurements of packet loss on the TCP layer is usually difficult because the TCP protocol handles packet loss and delivers successive prefixes of data to the application. Therefore, it is sometimes useful to fix a reasonable value for packet loss as input to the RA process instead. One implementation estimates that the losses are constant. Without any packet loss measurements, the RA can estimate the loss at 1%, or the RA can estimate the loss at 0.1%. The estimate may be determined by the type of connection, e.g., the estimate may be set at 0.1% for WiFi connection and 1% for cellular connection. Other methods, such as transforms in RTTs, can be used by RAs to infer packet loss indirectly. Alternatively, an implementation may obtain a packet loss estimate by querying the operating system for information about it.

Other implementations can estimate the loss in the application itself. To do this, it is assumed that the data from the network socket is generally received in maximum segment sized (MSS) chunks, but the packet loss is much larger chunks, a burst of approximately the size of the entire TCP receive window, Lt; RTI ID = 0.0 &gt; a &lt; / RTI &gt; Let M be the MSS in bytes (a good guess is M = 1500); Then, when n bytes are received, the number of transmitted packets is about n / M. z as the number of socket leads to cause a number of bytes read (read) than k · M, where k is some small integer. Assume k is chosen large enough that no more than k packets have arrived between the two network leads of the application. For applications that continue to wait on the socket, k = 3 would be nice. Then, p = z · M / n is an estimate of the packet loss probability. By computing z and n from the desired starting point, this procedure can estimate the packet loss rate over any desired range of time.

Given an estimate of RTT and a packet loss probability, the application can calculate the required good number of connections. The process can select a number of connections that are large enough to allow the target download rate to be achieved with that number of connections. The achievable rate of a single rate is generally limited by the TCP equation for the achievable throughput, which suggests that a single TCP connection can achieve an average download rate of T = MSS / (RTT -? P). Thus, the process may select the number of connections that are proportional to the target download rate divided by T.

The RA can also place lower and upper bounds on the number of TCP connections to use, for practical reasons. For example, an RA may limit the maximum number of connections it opens to 8 and the minimum number of connections to two.

Bandwidth, loss probability, and RTT are variable. The request accelerator monitors such quantities and dynamically changes the number of connections.

The request accelerator may split the HTTP requests into smaller sub-requests and reassemble the data responses returned for all sub-requests into a coherent response corresponding to the original request. There are many advantages to splitting requests into server requests. First, in order to use the available TCP connections, it is necessary to issue requests to all of them. The media streaming player may not provide sufficient requests to use all the connections. Request splitting mitigates these problems, since it results in a larger number of sub-requests, which can then be addressed on other connections. Second, request splitting yields shorter requests, which reduces the risk of data transmission at inappropriate times: if some TCP connections are temporarily slower than others, they can still be used as short requests. They will deliver a slower response than a fast connection, but the added relative delay to complete the entire request may not be so large in general because the requests are small.

In general, if more connections are used, it is desirable to create more subrequests per request. To achieve this, when there are n connections, the request accelerator may split each request into n sub-requests.

Other implementations choose the number of subrequests per request depending on the request size. If the subrequest size is chosen to be the size expected to take a fixed amount of time (say, 2 seconds) to download, the requests will be split into more subrequests if there are more connections to achieve the desired effect.

The partitioning rule should ensure that there are no unnecessarily small subrequests. For example, an RA implementation may force a minimum sub-request size in partitioning processes and may divide into fewer sub-requests if the minimum is not satisfied.

When multiple TCP connections are used, they probably compete for bandwidth. On a large time scale, each connection will receive the same amount as the others, but on a smaller scale, such as for a few seconds, some TCP connections may be significantly slower than others. This causes streaming problems, since it implies that some subrequests can take much longer than others, which can cause playback stalls.

To avoid this, the RA can use TCP flow control to "tame" connections. It can sufficiently limit the maximum receive window of each TCP connection so that no connection can use much more than its legitimate share of throttput. The amount of data (not yet acknowledged but transmitted) over the TCP connection is approximately the download rate divided by the RTT. Thus, if the TCP receive window is set at or slightly larger than the target download rate for the connection divided by the roughly estimated RTT, the download rate will be limited to approximately the target download rate or slightly larger. Thus, setting the TCP receive window size can be done as an administrator, ensuring that a given TCP connection will not download at a high rate that forces other TCP connections to download at a much lower rate. With such a mechanism in motion, connections tend to fetch at approximately the same rate, with slower connections having bandwidth available to speed up their fair share in that case, but at the same time, , Or a slightly higher aggregate download rate.

The RA can adjust the receiving window of the receiving client by adjusting the receiving buffer. It always resets these settings between consecutive requests.

One implementation may set the TCP receive window of each connection to be slightly larger than the product of the estimated RTT and the target download rate divided by the number of connections.

The target download rate may be determined from the media rate to be played back, for example. Other implementations may set the target rate based on the current playback rate (e.g., twice the current download rate).

6.1 RA Example

We now describe an embodiment of a request accelerator that includes the elements described above.

30 is a chart of operations for fetching into a plurality of TCP connections. Figures 30-31 illustrate operations under different conditions. For example, the connection to the Web server was bandwidth limited to 2 megabits per second ("mbps"), round trip time was 150ms, and packet loss was 0.1%. There were four active connections fetching fragments. The diagrams in Figures 30-31 represent the aggregate rates, RTT estimates obtained at the client, as well as the instantaneous rate of 4 connections.

In Figure 30, the receive buffers of the connections are not limited. In Fig. 31, they are limited to about twice the bandwidth-delay-product.

In the examples of Figs. 30 and 31, both methods stably achieve a full throughput of 2 mbps. 31), the transfer between connections is much more uniform: in most cases they are received at approximately the same rate. For connections with unconstrained windows (Fig. 30), this is not true at all, and in this case some connections are slower than others over long stretches.

Uneven connection rates are a problem for streaming applications because some urgent data is only being received very slowly (on a slow connection), while bandwidth is dedicated to faster connections that may fetch data that is not urgently needed This may mean that

The other difference between the unrestricted receive window and the restricted receive window is the RTT in which the client operates. If there is a limit, the RTT is kept close to the propagation delay and slow. Without the receiving window limit, the amount of data in flight exceeds the product of the base propagation delay and the connection capacity, resulting in a very prominent queuing delay and high RTT. High RTT is undesirable for media streaming clients because the client's response time to many events is typically a multiple of the RTT. For example, a client discovery event that causes new media content to be downloaded, or a client response time to a decrease in download speed that causes a cancel of a request or a change of representations, is generally a large multiple of the current RTT, The client's general sensitivity to RTT will degrade when the RTT is large.

32 is a flowchart of a request accelerator process.

Figure 33 shows a process for finding the number of subrequests to make for a given fragment request.

34 shows a process for selecting individual requests that are selected as being disassembly periods of source requests having computed sizes. In this process, the subrequest size is intentionally randomized such that the time during which connections are idle is different for each connection. This prevents all connections from being idle at the same time, thereby enabling better channel utilization. The request sizes are also sorted to help larger requests go ahead and limit differences in idle time.

Figure 35 shows a fragment structure for a repair segment determined by time offsets and time offsets.

In operation, the request accelerator receives HTTP requests from the SC (each request is in the URL and byte range).

The request accelerator downloads the requested byte ranges over HTTP and passes the data back to the SC when it is fully received. The RA achieves a sufficiently large download speed, but at the same time aims to ensure that each fragment is received before its deadline. High download speeds make it possible to select high quality video representations, while ensuring deadlines ensure that playback proceeds without stall.

To achieve the goal of high download speeds, the RA manages the changing number of open TCP connections, all of which are used to receive data over HTTP. The RA manages the details of how many connections to use, if necessary, to open or reopen them, and how to send requests to connections.

The RA will in some cases decide to split the source requests into smaller so-called RA requests that are then sent to different connections and response data that is transparently reassembled by the RA on arrival. For example, for a source request containing the first 64 kilobytes of a file, the RA can generate two RA requests, one 32 kilobytes chunk and the other a second 32 kilobytes chunk of the file. The RA can then request the two chunks in parallel on two different connections and generate a coherent 64 kilobyte response for the original request if both 32 kilobyte chunks have been received.

An RA can issue more RA requests than just the normal sub-ranges of source requests. For example, it may issue a request for FEC data of a fragment in addition to plain video data. In that case, the RA will transparently decode the FEC information if it has been received, and will only provide the final, decoded fragment to the source.

RA automatically tunes itself according to the network conditions. For example, if the RTT is large, the RA can decide to issue larger chunk requests to avoid much idle time between requests. Another example of automatic tuning is to keep the speeds of individual connections smaller so that the RA can guarantee the timeliness of those requests. To be able to do such, preferably the RA accesses the sockets of those connections directly. For example, in a Unix-like environment, it may be possible to set socket options using the setsockopt () function.

RA manages and keeps track of network conditions; This in particular involves measuring the download rate and the estimated round trip time (RTT). It collects this information because first, automatic tuning of the connection depends on their availability, and secondly bandwidth information needs to be passed to the SM, which uses it to calculate the rate estimates.

Another part of the information that the RA passes to the SM (via the SC) is the outgoing outstanding, ie, how much data of a given request has already been received. The SM uses the information for both the rate estimates and the request cancellation decisions.

The RA keeps track of the information needed to make bandwidth estimates by the SM. This information is the total amount of r-time spent on the download, T _r , and the total amount of downloaded bytes, Z. All of these numbers increase without decreasing, and are frequently examined by the SM. T _r timer is driven (if and only if) only when the at least one connection is active. A connection is considered active if it is sending an HTTP request or waiting for incoming response data. The Z counter counts incoming bytes and is aggregated across all connections.

RA download rate history

The request accelerator keeps track of a certain history of the rate by keeping the growing array of (T _r , Z) - pairs, which is stored in their respective order. We call this array mapTrZ. Updates to mapTrZ frequently occur at least at fixed time intervals (e.g., every 100 ms), and possibly also when new data is received.

RA uses mapTrZ to compute the windowed bandwidth estimate as follows. Consider the window of interest in width t, and let mapTrz [last] be the last entry in mapTrZ. Then mapTrZ [ i ]. T _r ≤ mapTrZ [last]. Find the largest index i that satisfies T _r -t . Note that i can be effectively searched by previous searches. The rate averages are then as shown in equation (8).

(식 8)

(Expression 8)

Equation 8 assumes that the difference in the subsequent Tr is small relative to t. This is ensured by sampling sufficiently frequently and never selecting a small window width t.

In practice, arbitrarily growing arrays are annoying. There is an upper bound on the maximum duration for which the past is being looked at, so there is a way to implement mapTrZ instead of a fixed-size ring buffer. This can be done as follows. The mapTrz array must be updated, and whenever the mapTrz array already contains at least two pairs, T _r - mapTrZ [last-1]. If T _r <100 ms, replace the last entry, or add a new entry.

6.1.2 Estimation of Round Trip Time ("RTT")

The RA collects bandwidth estimates. A priori, a simple way to obtain an RTT sample is to measure the difference in time at which an HTTP GET request is sent over the idle connection and the response begins to come in.

However, such measurements include queuing delays; If the client has other open active connections, then the link to the client has a lower rate than the rate at which it receives the data. The last hop to send data to the client can buffer multiple packets. In that case, the packets may be delivered with a longer delay than they do.

In our case, it is desirable to know the RTT discounting on the queuing delay induced by the activity of the client itself. To get an estimate of that amount, we proceed as follows.

During each period of activity, we collect RTT samples with the timing method described below; Each GET results in a sample. The current estimate is then the minimum of all of the samples. The list of samples is emptied each time the RA becomes inactive. (The client becomes inactive, for example, when the high watermark in section 3 is exceeded and downloads initiated are complete.) In the inactive periods, or in the active periods before any RTT samples are received, RTT The estimate is the estimate of the last notice.

The RTT estimator may also return an indication of &quot; no RTT estimate is unknown &quot;, which may be used, for example, at client initiation.

6.1.3 Tuning the Number of TCP Connections

Tuning TCP flow control allows the RA to keep the bandwidth on other connections approximately the same. Many configurable tuning constants are k _R (rate measurement window measured at RTT; suggested value: 30), k _N (proportional factor; suggested value: 8192 bytes), N _min ( N _target low cap); Suggested value: 1), and N _max ( N _target high cap; suggested value: 8).

The estimated bandwidth-delay-product (BDP ) is defined as BDP: = RTT · R , where RTT is the estimated RTT (as above) and R is the estimated (windowed method) The average reception rate during the last k _R RTT time.

The target number of connections is then defined as in Equation 9, where kN is a configurable constant.

(식 9)

(Equation 9)

The value of N _target is periodically recalculated. If the number of currently open connections is less than N _target , new connections are opened immediately to match N _target . On the other hand, if N _target is less than the number of currently open connections, no immediate action is taken. Instead, whenever the RA request completes, the RA checks to see if too many connections are open, and if so, closes the connection that just became idle.

6.1.4 Tuning TCP Receive Window on Connections

으로 설정한다. 여기서, c_w는 설정 가능한 하드코딩된 상수이고, 예를 들어, c_w = 3이다. RA는 접속의 TCP 수신 윈도우의 크기를 그것이 그 접속에 다음 HTTP 리퀘스트를 이슈할 할 때마다 설정한다.
RA sets the TCP receive window size for each connection

. Where c _w is a settable hard-coded constant, for example, c _w = 3. The RA sets the size of the TCP receive window of the connection whenever it issues the next HTTP request to that connection.

6.1.5 Request Split Process

Each source request delivered to the RA is potentially divided into more than one RA request, each of which corresponds to the different parts of the requested range. When all of the RA requests corresponding to a given source request are completed, the received data is reassembled into a complete fragment by the RA, which is then returned to the SC.

For a given HTTP request, the RA determines the number of RA requests, n , using a process that depends on some tunable values. The value of n depends on the following tunable constants: T _wn (rate estimate window width; proposed value: 4 s), D _min (minimum fetch duration; proposed value: 2 _s ), and c _s Proposed value: 6).

The process for finding the number n of sub-requests to be prepared for a predetermined fragment request is as shown in the pseudo code of FIG.

And the individual requests are selected as the disassociation period of the source requests using, for example, the process shown in FIG. 34, with the calculated sizes.

6.1.6 The Request Dispatch Process

The request accelerator maintains a set of RA requests. Whenever connections are ready to issue the next request, the request is dequeued from the RA queue if the queue is not empty and is addressed on an idle connection. If the queue is empty, a new fragment request is obtained from the SC. The request is then split into RA requests and queued on the RA queue. The queuing is preferably done in the order of the slices as returned by the process for finding the number of sub-requests to prepare a given fragment request.

HTTP connections may be shut down for various reasons, for example, because a web server timeout has occurred, or because a number of requests that may be issues on a single connection have been exceeded. The RA should handle this situation gracefully and transparently. Whenever a connection is shut down, the RA automatically reopens the connection. If the request was on a closed connection, it is dequeued from the connection, and a new RA request for the unreceived part is placed in front of the RA queue.

This procedure ensures that closed connections have minimal impact on performance.

6.1.7 RA Parameter Selection in Specific Examples

A TCP connection is limited by its flow control: an advertised receive window places an upper limit on the amount of data allowed to be unacknowledged at any point in time. Thus, if W represents the size of the receiving window and bdp represents the bandwidth-delay-product of the connection, we have bdp ≤ W (condition 1). The method in section 6.1.4 chooses the receive window size to satisfy this condition (1) if c _w > 1. This ensures that individual connections can not occupy substantially more than their legitimate portion of the available bandwidth. To allow for rate increases and to avoid a rate vicious circle, it is desirable to choose c _w that is somewhat larger than 1, for example, c _w = 2 or c _w = 4. The higher the value, the faster the rate grows but the connections are less fair to each other.

Other restrictions are imposed by the TCP congestion control process. If p represents the packet loss probability and M represents the TCP maximum segment size, then the rate r of a single connection is limited as indicated by Equation 10.

(식 10)

(Equation 10)

Now, it (using bdp = r · RTT and BDP = N · bdp) write with respect to the number of BDP and N connected again, we get to appear in the formula 11.

(식 11)

(Expression 11)

보다 조금 작게 선택되어야 한다는 것을 암시한다. M의 전형적인 값은 1kilobyte이고, 우리가 p = 0.01로 두면,

= 10 kilobytes이다. 따라서, 이 예에서, 식 9의 설정 N에 대해 섹션 6.1.3에서 제시된 바와 같이 k _N =8,192 bytes로 설정하는 것은 식 11의 부등식이 만족한다는 것을 보증한다. 수신기는 적절이 설정되고 프로그램될 수 있다.This _N k of the expression (9) to ensure that the steady-state where the inequality of Equation 11

Which should be selected to be slightly smaller. The typical value of M is 1kilobyte, and if we leave p = 0.01,

= 10 kilobytes. Thus, in this example, setting k _N = 8,192 bytes as set forth in section 6.1.3 for setting N in equation 9 guarantees that the inequality of equation 11 is satisfied. The receiver can be set and programmed appropriately.

We now look at the process in section 6.1.3 above to calculate the number n of RA requests for a given source request. A priori, we want to make slices as small as possible, because small slices provide many advantages; If a connection is slower than others, it will not cause problems with smaller requests, since smaller requests will complete faster on slower connections. Thus, in a small slice configuration, a slow connection will eventually serve essentially only fewer requests. Another advantage of the small slices is that they tend to involve relatively short sections of the buffer in time, and thus tend to incorporate the effort into the most urgent work areas.

However, making the slice small costs money: First, each request includes some overhead on both the uplink and downlink. Second, after completing a request, the connections will be idle for about RTT. Therefore, the request division process should ideally select chunks as small as possible under the condition that they will not excessively cause uplink traffic and fully utilize the capacity of each available link. The preferred characteristics are therefore:

1. D _{min It} targets one request per connection per real time. This ensures that the uplink traffic is limited by a value proportional to N _target in the worst case.

2. c _s · At most one request per connection per RTT . This makes the active time of the connection at least about c _s / ( c _s + 1), i.e. close to 1 for moderate c _s .

A good choice of D _min depends on the application. It is usually the normal duration of a fragment to select it as approximately the desired end-to-end delay (but smaller). If the end-to-end delay is large, larger buffers can be used, and the worse impact of larger slices is smaller. On the other hand, for short end-to-end delays, the buffers are small, and therefore the slices must be small to avoid slow connections that cause stalls. In that scenario, the higher cost of the smaller request becomes the value of the stability obtained at the buffer level.

The parameters used may be tuned according to the profile indicator of the MPD (Media Presentation Description), which is a summary of the characteristics of the media streamed to the client. Instead of downloading all media segments and showing them to the end user, the client can choose to &quot; skip &quot; segments based on different usage events from the profile in the MPD.

The lower bound of the c _s selection can be devised as follows. If there are open N connections and the RA is active, there will be approximately N c _s / ( c _s + 1) active connections on average. To ensure that the receive window for all N connections to a larger sum ever enough to sustain the total target rate, is preferably at least _{1 c w · c s / (} c s + 1).

This limit is conservative. The estimated number N · c _s / ( c _s + 1) of active connections is only average and does not take deviations into account, although there may be some variation. In practice, it is advantageous to make c _s 2 to 3 times the values suggested by the above limits, for example c _w = 3 and c _s = 6, c _w c _s / ( c _s + 1) is at least 2.5.

6.2 RA with forward error correction

If data is received over multiple TCP connections, they sometimes have different download rates temporarily. If the request of a fragment is divided into several sub-requests, the entire fragment is received only when the last sub-request response (chunk) is received. This is a problem when a fragment needs to be urgently received because one of the sub-requests may be processed on a slow connection to prevent the fragment from being received quickly.

In addition to the video data, the content provider may provide additional forward error correction ("FEC") repair data for each fragment that the client may fetch to help rebuild the original fragment. For example, suppose a client needs to have 4 connections and receive a fragment of 4000 bytes in size urgently. The request accelerator will split the fragment into four ranges of 1000 bytes each and issue one request for each of the four connections. Connection 1 is fast and connection 4 is moderately fast, but the second and third connections may be very slow. Thus, even though the overall download rate is in principle high enough to timely download the entire fragment, it can only arrive very slowly because connections 2 and 3 are blocked.

To avoid this problem, the client may attempt to fetch the same data as it would like to connect 2 or 3 using connection 1 as soon as it processes its own sub-request. This can be helpful, but the RA will tell you which connections need more help; You must decide whether it is connection 2 or 3. If it makes a false prediction, it is downloading redundant data without the need, and fragments may still not arrive in time.

Instead, a better request accelerator can fetch some repair data using connection 1. The repaired (i.e., FEC coded) data may be used to reconstruct the lost data in response to whether data from request 2 or 3 is lost, if successfully downloaded. The only constraint is that the amount of data received is sufficient to reconstruct the fragment. In other words, in our example, the number of repair bytes plus the number of received fragment bytes should be greater than or equal to 4000.

In one implementation, the content provider provides access to FEC repair data for the coded video segments. It makes repair data available in a manner similar to the original video data. For example, it may provide, for each media segment file, an additional FEC file containing repair information. The content provider may provide the necessary information and parameters to use the FEC of the media presentation description. In other implementations, the media presentation description does not include any information about the FEC, but the client can access it using a common convention, such as how to infer the name of the FEC repair URL from the segment URL.

The client implementation implements processes for how and when to request the repair data. The amount of requested repair data may depend on how much data is out-standing. It can also depend on how quickly the fragment needs to be used. For example, if there is enough time left, you will want to receive all the source data in a timely manner, and it is probably unnecessary to request arbitrary repairs. On the other hand, if the fragments are urgent, the stall is imminent and the client may not be able to obtain sufficient data for the fragment in a timely manner, so it will want to request a lot of repair data. Thus, an implementation may set the amount of requested repair data to ? (B) S , where S is the amount of outgoing source data and ? (B) is a decreasing function of the buffer level.

Other implementations can set the amount of out-standing data to the amount of out-standing data that is proportional to the amount of out-standing data of the most incomplete request, rather than the total out-standing amount.

6.2.1 Example of a Water Segment Generator

All of the following calculations, which may be related to how the DASH standard uses FEC, in particular RaptorQ for FEC, is preferably performed using fixed-point / integer operations. This involves calculating the number and positions of the source symbols in the fragment of the representation, and calculating the number and positions of the repetition symbols for the fragments in the repair segment should be performed using fixed-point arithmetic. This needs to be accomplished by an ingestion process in which exactly the same result as the RA process of decoding source fragments using combinations of received FEC repair fragments and source fragments produces FEC repair fragments from the source segments , So these calculations must have exactly the same result.

Using floating-point arithmetic instead of fixed-point arithmetic can result in subtle buggy behavior that is sometimes hard to find due to the different corner case behavior of different floating-point implementations on different platforms, - Points will not be accepted in the standard that should yield exactly the same calculation results.

The repair segments may be generated in a separate process based on the already processed source segments including the sidx tables. In addition to the source segments themselves, the two inputs to the process are the waterline ratio R and the symbol size S. To facilitate the use of fixed-point arithmetic for the calculation of the number and positions of the repair symbols in a segment in a segment, the R value may be expressed in units of mille, i.e. R = 500 means that the ratio is 1/2 do.

Within each segment, at the beginning of the source segment, there is segment indexing information, which contains a time / byte-offset segment map. The time / byte-offset segment map includes time / byte-offset pairs T (0), B (0), T (1), B (1) ... , ( T ( i ), B ( i )), ... , A list of (T (n), B ( n)), where T (i -1) is started in the segment for the playback of all media, the i-th fragment of a segment of the media involved in the initial starting time of the media time It represents a, T (i) is the end time of the i-th fragment represents a (thus the start time for the next fragment), byte-offset B (i -1) is the i-th fragment beginning of the media associated with the beginning of the source segment and a source corresponding to the start byte index of the data in the segment, B (i) is the number of bytes corresponding to a segment to the i th fragment and stores them (so B (i)) is a first fragment of i +1 Byte index). If the segment includes a plurality of media components, T ( i ) and B ( i ) may be provided to each component of the segment in an absolute way, or they may be displayed relative to other media components serving the reference media component . In any case, B (0) is the starting byte index of the first fragment of the segment, which may be greater than zero due to the sidx information preceding the first fragment of the segment. If B (0) is not zero, there are some water symbols at the beginning of the water segment corresponding to sidx. Depending on the implementation, these first repair symbols may protect the data of the segment until the beginning of the first fragment, or they may be unused padded-zero data bytes.

The repair rate R may be signaled to the MPD along with the repair segment metadata, or may be obtained by other means (TBD). As an example of a value for R, if R = 500, then the waterline segment size approximates (very closely) 0.5 times the corresponding size of the source segment generated therefrom, and the size of the waterline fragment of the waterline segment corresponding to the source fragment in the source segment is It also approximates (very loosely) 0.5 times the size of the source segment. For example, if the source segment contains 1,000 kilobytes of data, the corresponding line segment contains approximately 500 kilobytes of repair data.

The S value may also be signaled to the MPD along with the waterline segment data, or may be obtained by other means. For example, S = 64 indicates that the source data and the repair data contain symbols of size 64 bytes each for coaming and decoding where FEC is FEC. The S value may be chosen to be proportional to the streaming rate of the representation of the associated source segment. For example, if the streaming rate is 100 kbps, S = 12 bytes may be appropriate, whereas if the streaming rate is 1 Mbps, S = 120 bytes may be appropriate and if the streaming rate is 10 Mbps, S = 1,200 bytes may be appropriate . One goal may be how to divide the granular fragments into symbols and have a good trade-off between the processing conditions for FEC decoding versus the streaming rate. For example, at a streaming rate of 1 Mbps and fragments near 500 ms in size, each fragment is near 64 KB of data, and if S = 120, the fragment consists of approximately 500 source symbols, Of data is needed to recover the source blocks, which means that the extra reception required due to the symbol granularity is up to 0.2% times the number of HTTP connections for which the fragment is being received. For example, if the number of HTTP connections is 6, the symbol granularity reception overhead is limited to 1.2%.

A received segment may be generated for the source segment as follows. Each fragment of the source segment is considered a source block for FEC encoding purposes, and thus each fragment is treated as a sequence of source symbols of the source block from which the repair symbols are generated. The total number of repetition symbols generated for the first i-th fragment is calculated as TNRS (i) = divceil (R * B (i), S * 1000), where divceil (I, J) Divxeil (I, J) = (I + J-1) div J, where div is a fixed-decimal division in which the result is rounded down to the nearest integer. Therefore, the number of repair symbols generated for fragment i is NRS (i) = TNRS (i) - TNRS (i-1).

The repetition segment includes concatenation of the repetition symbols for the fragments, the order of the repetition symbols in the repetition segment is the order of the fragments from which they are generated, and the repetition symbols in the fragment are the symbols of their encoding symbol identifiers (ESI & Order.

By defining the number of repair symbols for a fragment as described above, the total number of repair symbols for all previous fragments, and therefore the byte index and byte range for the symbols of the repair fragment i, are only R, S, B (I- 1) and B (i), and does not depend on any previous or subsequent structure of the fragments in the source segment. This makes it possible for the client to quickly calculate the starting point of the repair block in the repair segment and to quickly calculate the number of repair symbols in the repair block, using only local information on the structure of the corresponding fragment of the source segment from which the repair block is generated So it is beneficial. Thus, if a client decides to start downloading and playing a fragment from the middle of the source segment, it can also quickly create and access a corresponding repair block corresponding to the fragment from within the corresponding repair segment.

The number of source symbols of the source block corresponding to the fragment i is calculated by NSS (i) = divceil (B (i) - B (i -1), S). If the B (i) -B (i -1) is not a multiple of S, for the purposes of FEC encoding and decoding, the last symbol is padded out with 0 bytes, i, So that its size is S bytes for FEC encoding and decoding purposes, but these 0 padding bytes are not stored as part of the source segment. In this embodiment, the ESIs for the source symbol are 0, 1, ... , NSS (i) -1, the ESIs for the repair symbols are NSS (i), ... , NSS (i) + NRS (i) -1.

In this embodiment, the URL for the repair segment may be generated from the URL for the corresponding source segment, for example by simply adding a suffix &quot; .repair &quot; to the URL of the source segment.

The waterline segment may also be part of the corresponding source segment, for example, attached at the end. The structure of the combined segment may also be that the source fragments and the repair fragments are contiguous in the combined segment, i. E., The combined segment is the first source fragment, the following first repair fragment, the following second source fragment, A second repair fragment, and the like. Those skilled in the art will appreciate that the above-described methods and processes may be readily adapted to apply to such coupled segments.

6.2.2 Implementation of a Request Accelerator Using Repair Segments

The repair indexing information and the FEC information for the repair segment are implicitly defined by the indexing information for the corresponding source segment and implicitly from the R and S values, where R is expressed as an integer between 0 and 1000 representing mille, and S Is represented by bytes. The fragment structure and time offsets including the repair segment are defined by the time offsets and structure of the corresponding source segment. The byte offsets for the start and end of the repair symbols of the repaired segment corresponding to fragment i are RB (il) = S * divceil (R * B (il), S * 1000) and RB B (i), S * 1000). (I) - RB (il), so the number of the waterline symbols corresponding to the fragment i is NRS (i) = RB (i) - RB / S. &Lt; / RTI &gt; (Note that the divceil operation is not needed here, since the numerator is guaranteed to be a multiple of S, but divceil can be used here and the result will still be correct.) The number of source symbols corresponding to fragment i is NSS (i) = divceil B (i) -B (il), S), where the last source symbol is padded with zeros if necessary for decoding purposes, as described for encoding. Thus, the repair index information and corresponding FEC information for the repair block in the repair segment can be implicitly derived from R, S and indexing information for the corresponding fragment of the corresponding source segment.

As an example, consider the example shown in FIG. 35, which shows fragment 2 starting at byte offset B (1) = 6,410 and ending at byte offset B (2) = 6,770, i.e., fragment 2 is 6,770-6,410 bytes size And 6,770 is the start byte index of Fragment 3. In this example, the symbol size is S = 64 bytes and the vertical lines of the dashed line represent byte offsets in the source segment corresponding to multiples of S. The total contour segment size as a percentage of the source segment size is set to R = 500 per mille in this example (the waterline is approximately one half of the source). The number of source symbols of the source block for Fragment 2 is calculated by NSS (2) = divceil (6,770-6,410,64) = (6,770-6,410 + 64-1) div64 = 6, 0, 5, where the first source symbol is the first 64 bytes of Fragment 2 starting at byte index 6,410 in the source segment and the second source symbol is the next 64 bytes of Fragment 2 beginning at byte index 6,474 in the source segment, And so on. The end byte offset of the repair block corresponding to Fragment 2 is calculated as RB (2) = 64 * divceil (500 * 6,770,64 * 1,000) = 64 * (3,385,000 + 64,000-1) div 64,000 = 64 * 53 = 3,392 , The start byte offset of the repair block corresponding to the fragment 2 is calculated as RB (1) = 64 * divceil (500 * 6, 410, 64 * 1,000) = 64 * (3,205,000 + 64,000-1) div 64,000 = 64 * 51 = And thus in this example are the two waterline symbols of the waterline block corresponding to Fragment 2 with ESIs 6 and 7 starting at byte offset 3,264 and ending at byte offset 3,392, respectively, in the waterline segment.

This is shown in Fig. In the example shown in FIG. 35, even if R = 500 (the waterline is approximately one half of the source) and there are six source symbols corresponding to Fragment 2, if we simply used the number of source symbols to calculate the number of waterline symbols , As expected, the number of repair symbols is not 3, but is instead calculated to be 2. In contrast to simply using the number of source symbols in a fragment to determine the number of repair symbols, the method performed here can only calculate the position of the repair block in the repair segment from the index information associated with the corresponding source block of the corresponding source segment Let's do it. It is important to note that in order for this to be a computation consistent with the ingestion process and to be in the RA process, the number and positions of the repair symbols for the repair fragments in the repair segment are calculated using fixed-point arithmetic. Furthermore, as the number of source symbols in the source block, K, increases, the number of the repair symbols in the corresponding repair block, KR, is closely approximated by K * R / 1,000 and is generally divceil (K * R, 1,000) , KR is at least divfloor ((K-1) * R, 1000), where where divfloor (I, J) = I div J.

Illustrated Examples

25 shows a rate selection process. The larger the settings for λ and μ, the more aggressive the settings are. Fig. 23 shows other values for the parameter lambda. Fig. 24 shows different values for the parameter [mu]. Hybrid configuration seeks to monitor rate fluctuations by two major mechanisms. The first is to be more careful to increase the rate when B is bigger, and the next is to work harder to stay at the current rate when B is smaller.

An exemplary setting for pker xy: C = x * min (y * Tdl, B) may be set to xy values of 8.1, 4.2, 2.4, 4.4, or other xy values. Note that the actual average window of pker is longer than C due to the skip of the download pending cycle. Assume that the EWMA does not skip and the rate in the download pending period equals that of the last download period.

For MWA (Moving Window Average), H (z) = (1 / D) * ((1-z ^D ) / (1-z ^-1 )) and D is the window size. X _i = min {R _k : _k ≥ i}, where R _k is the EWMA of the rate with the weight W _k is RH, W ₁ <W ₂ <W ₃ <... to be. For EWMA, H (z) = ((1 -?) / (1 -? Z ^-1 )) and? Is the weight of the previous average. MWA and EWMA are approximately equal in some cases.

If the adaptivity estimator has a long average window, it reduces the rate switch frequency and at the same time maintains approximately the same average rate for live streaming. Other settings work well for different scenarios. Aggressive configurations work well for more stable scenarios, while less aggressive configurations are better suited for more volatile scenarios. If the bandwidth is higher than the highest representation rate by any margin over a significant fraction of the time (e.g., the% of time at 20-sec average is higher than the rate cap) It is beneficial to go. Ideally, the device should be able to detect scenario types and apply appropriate settings. Scenario detection may be based on factors such as the Lyoto technology type, the number of rate variations within a unit time, the speed of movement, and so on. A simpler strategy can be based on the above observations: Use a more aggressive setting when the &quot; total &quot; bandwidth is higher than the rate cap.

8. Setting of rate selection parameters

In this section, examples are provided for setting rate selection parameters.

For MLB, EFF = 1 - Rv / Rdl, Rv is the current rate of the selected representation and Rdl is the current download rate. The proposed rule is as follows:

If EFF <0, it probably drops more than one rate

If 0 < = EFF < 0.1,

If 0.1 <= EFF <0.6, the current rate

If 0.6 < = EFF < 0.8,

If 0.8 < = EFF < = 1,

Let alpha = Rv / Rdl. Then it changes roughly to:

If alpha <= 0.4, then at least one rate rise

0.4 < alpha < = 0.9,

If 0.9 < alpha, then at least one rate drop

Put it in the context of the DASH client rate selection process:

Let RUP be the rate of representation corresponding to UP, RDOW be the rate of representation corresponding to DOWN, and let Rv be the rate of the currently selected representation as above. RUP is selected as large as possible to be RUP <= lambda (t) * Rdl, and RDOWN is selected as large as possible to be RDOWN <= mu (t) * Rdl. Where B is the current amount of presentation of the media buffer, D is the time limit to the next possible switch point beyond the point at which it is currently being determined, and delta is the network latency &lt; RTI ID = And round trip time, for example, delta may be set to one or two seconds as an approximation, or delta may be set according to the measured upper limit of the current RTT.

The overall selection of the following rate RNEXT is as follows:

If RUP <Rv, RNEXT = min {Rv, RDOWN} or RNEXT = RUP.

The above MLB parameters may be approximated by setting lambda (t) = 0.4 * R and mu (t) = 0.9 for all t, where R is the rate of the current representation of the rate of the next higher representation Ratio. For example, if the current rate is 500 Kbps and the next higher rate is 750 Kbps then R = 1.5 and thus lambda (t) = 0.6. This approximates the MLB process as follows.

At the decision point, if EFF > = 0.6, i.e. alpha <= 0.4, then Rv <= 0.4 * Rdl and in this case RUP will be at least Rv * R (since lambda (t) = 0.4 * R for all t ) Thus, RNEXT = RUP, i.e. the rate can rise to the next higher representation of Rv * R, and if Rdl is much larger than 0.4 * Rv then RUP will be larger than Rv * R Lt; / RTI &gt; and EFF is greater than 0.8, for example, RUP will be greater than one rate above Rv * R. If EFF <0.1 then Rv> 0.9 * Rdl, then RDOWN will be less than Rv (because RDOWN <= 0.9 * Rdl) and the rate will drop, ie RNEXT <Rv. If EFF is between 0.1 and 0.6, then RUP <= Rv * R and RDOWN> = Rv, then RNEXT will be selected equal to Rv.

Rate selection parameter sets

The following tables specify several possible rate selection parameter sets. The values of lambda and mu for the median of t that are not shown in the table below should be calculated by interpolating linearly between the surrounding values. The values of lambda and mu for the value of t above those shown in the table below shall be set to the lambda and mu values for the maximum value of t indicated.

If the constraints mu (t) <= t and lambda (t) <= t are satisfied for all t, theoretically there will be no stalls in playback, but a small stall in playback It may be desirable to continue playing at a much lower rate, for example, a jump from 1 Mbps to 20 kbps may be a worse experience than a jump from 1 Mbps to 250 kbps with a one second pause in between. Note that the minimum values of lambda and mu are set in the tables of FIG. 36 and a stall will occur for mu (t)> t and / or lambda (t)> t (nonetheless lambda And mu (t), the stall may occur in any case where the buffer is so busy).

As will now be described, the client device may provide rate adaptation and download processes for adaptive video streaming over HTTP. Clients streaming video over the Internet (and other networks) are confronted with the problem of fluctuating bandwidth. When high quality video is streamed, the link may not be fast enough from time to time, which causes the player to stop and buffer again. In other cases, low quality video is a much worse user experience, with much lower bandwidth. One solution is to adaptively adjust video quality: choose a higher quality when the throw foot is high, and automatically switch it down.

Adaptive video streaming, however, causes a number of challenges: (1) The process or algorithm for selecting a video rate (quality) must act fast enough to accommodate rate reductions as well as rate increases. At the same time it avoids too early or strange decisions and avoids unnecessary rate switching decisions. The client should aim to fetch data at a rate high enough to allow high video quality to be obtained. At the same time, the download process must ensure that the data is received in a timely manner. Each frame must be received entirely before it is played out. They should be able to achieve these goals without requiring an unnecessarily large playback buffer. Some problems with large buffers are that, for live events, the amount of video in the buffer is limited by target-to-target latency, which severely limits the playback buffer available in these cases. Also, relying on a large buffer causes undesirable delays in playback starts and searches because the buffer needs to be pre-populated. Also, the large playback buffer uses a lot of memory, which is not enough for mobile phones and other client devices.

A process for rate estimation that will quickly respond to new rate changes to solve these problems. Rate estimation is an adaptive windowed average that is particularly well suited for use in streaming video. The rate estimator takes into account changes in the video buffer level and the video buffer level in such a way as to be able to ensure that the rate adapts fast enough, if necessary, while keeping the window wing width large (and hence larger measurement variance). The assurance provided by the process is: (a) if B is the amount of video data in the buffer (in seconds of playback time) when a rate drop occurs, then the estimator estimates the rate within the time it takes for the buffer to fall to B / 2 (B) if B is the amount of data in the buffer during which the rate increase occurs, the rate estimator adjusts to the new rate fast enough so that it can appear in principle at most 3 * B (assuming a smart rate change process) .

The rate determination process may (a) use the buffer to avoid changing the rates abnormally, even when low download rate estimates are observed, and c) In a stable rate scenario, quickly select the right stable rate. (a) to take accurate estimates, (b) to achieve link capacity even at high network delays and packet loss rates, and (c) to use multimedia download strategies for HTTP to achieve timely delivery of streams. To achieve this, it is possible to use multiple HTTP connections, decompose media requests into smaller chunked requests according to network conditions, synchronize connections using TCP flow control mechanisms, and send data of bursts to requests can do. We can also use the HTTP pipelining process to keep the connection busy.

Many features, aspects, and details have been described so far. As described, the method steps in various embodiments may be performed by corresponding programmed elements, instructions provided to the processor, hardware, or other devices, as would be apparent to those skilled in the art. Similarly, elements may function by processes or program elements. The structure of the elements of an embodiment may simply include a set of instructions that are executed by the processor but have been described herein as corresponding method steps.

In various embodiments, a download rate acceleration may or may not be used. One example of a download rate change is a method or device that uses HTTP requests over TCP connections to speed up downloads. A TCP connection has a specific window size and the nodes at the ends of the TCP connection can change settings for the window size. Setting the window size for successive HTTP requests whose size is a function of the target download rate is novel (on novelty). Thus, as the target download rate changes, the TCP window size may change.

In one embodiment, a method and / or apparatus or computer readable medium for controlling the downloading of data over a network path between a source and a receiver coupled by a network path is used, the method comprising: The TCP connection between the source and the receiver may be a direct connection or an indirect connection and may determine a target download rate for the media content, Rate includes varying between at least two values for at least two consecutive HTTP requests and downloading media data elements of the media content to be downloaded using each TCP connection of the plurality of TCP connections, It is part or all of the response of the requests, The determined TCP receiver window size for the CP connection is determined based at least in part on the target download rate and the determined TCP receiver window size varies between at least two values for at least two consecutive HTTP requests.

The determined TCP receiver window size for the current TCP connection is determined based at least in part on the product of the current estimated round-trip time (&quot; ERTT &quot;) for the current TCP connection multiplied by the multiplication rate, The target download rate for the TCP connection, and a rate that is higher by a predetermined amount than the target download rate. The current ERTT can be determined by measuring the minimum observed RTT for the immediately preceding measurement period, such as 1 second, 10 seconds, 50 seconds, and so on. The current ERTT can be determined by measurement at the end of the period during the outage, while the period of outage is after the download period and no active HTTP requests are presented over TCP connections for a pre-determined duration period. The target download rate may be proportional to the current aggregate download rate over all used TCP connections, divided by the number of TCP connections used, such as twice or three times the download rate of the current total. The target download rate may be proportional to the playback rate of the media content and the playback rate is the rate through the set across all used TCP connections divided by the number of TCP connections used. Each media data element may be partitioned into a number of chunks having sizes within a predetermined range of variances, and the number of such chunks is based on the number of TCP connections used. The n of such chunks may be further based on at least one of the estimated round-trip time (&quot; ERTT &quot;) for current TCP connections, the current download rate, and / or the size of the requested media fragment. The predetermined range of deviations can be zero, so each chunk has the same size per fragment request, and the number of chunks is a predetermined number of times the number of TCP connections used. A later HTTP request for a subsequent media data element may be assigned to the first available TCP connection.

The control determining the number of TCP connections to use between the source and the receiver, the number being greater than 1, the number of TCP connections to use being determined based at least in part on at least one determined network state, Downloading a plurality of media data elements of the media content to be downloaded using each of the media content, wherein the media content is part or all of a response to the plurality of HTTP requests. The number of TCP connections used may be based on an estimate of the round-trip time (&quot; ERTT &quot;), target download rate and loss rate for TCP connections. The loss rate may be estimated at 1% or 0.1%. The number of TCT connections to be used may be between 2 and 16, inclusive and / or proportional to (a) the target download rate, (b) ERTT, (c) the product of the square root of the estimated loss rate. For each TCP connection, the TCP receiver window size can be determined based on the target download rate for that TCP connection, and the determined TCP receiver window size varies between at least two values for at least two consecutive HTTP requests.

In one embodiment, a method and / or apparatus for download estimation, which takes into account the presentation buffer and estimates the download rate based on how large / full / empty the buffer is, Is used. For example, estimating a download rate at a receiver coupled to data sources by a network path having a finite bandwidth, the download rate being the rate at which data can be received over the network path at the receiver, And the presentation buffer stores media data between at least the time at which the media data is received and the time at which the media data is consumed by the presentation element associated with the receiver, and wherein the estimate of the download rate is non- The buffer level at a given time is at least approximately equal to the amount of media data that has been received at that time but has not yet been consumed by the presentation element On Corresponding to that occupied by, and can comprise the application as part of the measurement of the downloaded characters are stored estimated rate display.

The presentation element may include a display and an audio output. The estimation period may have a proportional duration, with a predetermined proportional factor, at the measured buffer level. The duration of the estimation period is proportional to the number of bytes of uncommitted media data in the presentation buffer at the measurement time and / or is a function of the additional rate at which the media is added to the presentation buffer, and / And may be proportional to the time used to download the determined portion. The predetermined duration may correspond to a duration for which a predetermined portion of the content of the presentation buffer has been downloaded. The estimation period may be the smaller of the time at which a predetermined portion of the content of the presentation buffer was downloaded and the presentation time of the media data in the presentation buffer.

In one embodiment, a method and / or apparatus or computer readable medium for selecting a playback rate is used, the playback rate being the rate at which the media is consumed from the presentation buffer, in megabits / second, . If the receiver requests for some media, there is a playback rate for that media. Frequently, but perhaps not always, higher quality media has a higher playback rate and thus shows a trade-off. Which playback rate to use / request is a function of how much media is in the presentation buffer, at least occasionally. The receiver may receive the media to play out using the presentation element of the receiver and the playout may cause the media to be consumed at the playback rate from the presentation buffer and the receiver to select from the plurality of playback rates And wherein the presentation buffer stores media data between at least the time at which the media data is received and the time at which the media data is consumed by the presentation element associated with the receiver and stores an indication of the buffer level And the buffer level corresponds to how many presentation buffers have been received by the media data that have been received but not yet consumed by the presentation element, determine the estimated download rate, and store the stored representation and the estimated download rate To calculate a target playback rate and selecting from among a plurality of playback rates in accordance with a target playback rate.

The selected playback rate is less than or equal to a predetermined multiplication of the estimated download rate, and the predetermined multiplication is an increasing function of the buffer level. The predetermined multiplication may be an affine linear function of the playback duration of the media data of the presentation buffer and the predetermined multiplication may be less than 1 when the buffer level of the presentation buffer is less than the threshold amount. The predetermined multiplication may be greater than or equal to 1 when the presentation duration of the media data in the presentation buffer is greater than or equal to a preset maximum amount of presentation time. The predetermined multiplication may be a piecewise linear function of the playback duration of the media in the presentation buffer. The selected playback rate may be less than or equal to a predetermined multiplication of the estimated download rate and the predetermined multiplication may be an increasing function of the number of bytes of media data in the presentation buffer. The playback rate may be selected to be the largest available playback rate among a plurality of playback rates that is less than or equal to the proportional factor multiplication download rate estimate and the proportional factor may be a presentation divided by an estimate of the response time to rate changes Is an increasing function of the playback duration of the media data in the buffer. The response time may be upper bound to the presentation time between switch points of the media data and / or the estimate of the response time may be an average of the presentation time between switch points of the media data. The estimate of the reaction time may be greater than or equal to a predetermined constant times the estimated round-trip time (&quot; ERTT &quot;).

A receiver for receiving media to playout using a presentation element of a receiver, the playout resulting in media being consumed at a playback rate from a presentation buffer, and the receiver being configured to select from a plurality of playback rates , The receiver monitors the presentation buffer and the presentation buffer stores media data between at least the time at which the media data is received and the time at which the media data is consumed by the presentation element associated with the receiver, And the buffer level corresponds to how many presentation buffers have been received by the media data that have been received but not yet consumed by the presentation element, And a method or apparatus for calculating a rate of layback and selecting among a plurality of playback rates in accordance with a target playback rate.

The playback rate may be selected based on an estimate of the response rate for high proportional factors, low proportional factors, download rate estimates, current playback rate, buffer level, and rate changes. Both the high proportional and low proportional factors may be incremental and / or delimited linear functions of the playback duration of the media data in the presentation buffer divided by an estimate of the response time to the rate change, May be greater than or equal to the lower proportional factor. The playback rate may be equal to the previous playback rate if the previous playback rate is between a low proportional factor multiplied estimated download rate and a high proportional factor multiplied download rate estimate. If the previous play rate exceeds the high proportional factor multiplication download rate estimate, then the playback rate can be selected to be the highest available factor multiplication multiplied by the largest available play rate that is not greater than the estimated download rate. If the previous playback rate is below the low proportional factor multiplication download rate estimate, the playback rate may be selected to be the largest available playback rate that is not less than the low proportional factor multiplication download rate estimate.

In one embodiment, a method and / or apparatus or computer readable medium is used to determine whether to receive requests and to cancel requests in the process. It is desirable for the receiver to receive requests for segments / fragments / fragments of media, receive responses to requests, store media from responses and possibly other requests, it is desirable to cancel the request and issue another request Can be determined. The playback rate of the media can be determined by choosing the highest playback rate that the receiver expects to acquire the most aggressive and the media in the presentation buffer as it is consumed. If the download rate suddenly drops, the receiver cancels the current request and decides whether to make a new request for low playback rate media or the current request to play out. Canceling a high playback rate request and replacing it with a low playback rate request may cause the content of the presentation buffer to be long-lasting, but canceling the request on the way may result in a failure of any partially received media It can cause loss.

In such an embodiment, the receiver receives the media to playout using the presentation elements of the receiver, and the playout results in the media being consumed at the playback rate from the presentation buffer, and the receiver has a plurality of playback rates As shown in FIG. Determining a request behavior includes monitoring the presentation buffer and storing the media data between a time at which the media data is received and a time at which the media data is consumed by the presentation element associated with the receiver, And the buffer level corresponds to how many presentation buffers have been received but has been occupied by media data that has not yet been consumed by the presentation element and the number of issues requested to download the selected first chunk of media data And if the issued request is out-standing, determining whether to continue the request or to cancel the request, based on the network status and the status of the issued request.

Determining whether to continue the request or to cancel the request includes determining whether there is enough time to complete the download to the request before the first media data is played out and canceling the request if there is not enough time . Determining whether to continue the request or to cancel the request will determine that a stall will occur based on the download rates and the media consumption rates and will cause the presentation element to consume media data at the rate indicated by the media being consumed Estimates a stall period between the time when the media element is not available and the time at which the presentation element can resume consuming the media data at the rate indicated by the media being consumed and determines the effect of the continuation or cancellation on the stall period And canceling the request may further include canceling the request if the stall period is reduced.

Other features select a second chunk of media data, the second chunk of media data has a start presentation time, the start presentation time is a start presentation time same as the first chunk of media data, Requesting download of two chunks, selecting a second chunk of media data, the second chunk of media data having a start presentation time, the start presentation time of which is later than the first chunk of media data, And requesting download of the second chunk. The second chunk of media data is the lowest chunk available to the receiver as compared to that of the start presentation time of the first chunk by the receiver and / May be selected to be the maximum playback rate with a predetermined maximum gap between the start presentation time of the first chunk of data.

Some embodiments may determine whether the download of the remainder of the first chunk of media data can not be completed in time for playback and determine whether the download of the second chunk of media data can be completed in time for playback Determine whether to continue the request, or cancel the request for the first chunk of media data and instead request the second chunk of media data, if the download of the remainder of the first chunk of media data is a playback , And based on whether the download of the second chunk of media data can be completed in time for playback. The playback rate of the media data of the second chunk of data may be selected to be the highest playback rate supported by the receiver. The receiver may request media data that already covers the presentation time of at least some of the media data in the presentation buffer, download the requested media data, play out the requested media data, At least some of the media data to be played back may be discarded. The playback rate of the requested media data may be a maximum playback rate, subject to constraints on the maximum presentation duration of the corresponding media data discarded from the presentation buffer. The requested media data may be selected such that its start presentation time is the earliest start presentation time available to the receiver.

In some receivers, the download depends on the buffer level, and receivers use the concept of high watermark and low watermark. In such a receiver, the media data is downloaded from the source and stored in the presentation buffer of the receiver. The filled level (or simply &quot; level &quot;) of the presentation buffer is determined, and the filled level indicates that some portion of the presentation buffer contains media data that has not yet been consumed by the presentation element. If the filled level exceeds the high fill threshold (&quot; high watermark &quot;), the download is stopped and if the filled level is below the low fill threshold (&quot; low watermark &quot;), the download resumes. The filled level can be updated as the media data is consumed by the presentation element. The filled level can be measured in units of memory storage capacity and / or presentation time units. The download may be based on an estimated round-trip time (&quot; ERTT &quot;) and the ERTT is reset when the media data download is restarted. If the download occurs over a plurality of TCP connections, the plurality of TCP connections used may be reset when the media data download is restarted. High fill and low fill thresholds can change over time.

Additional embodiments may be envisioned by those skilled in the art after reading this disclosure. In other embodiments, combinations and sub-combinations of the inventions disclosed above may be made advantageous. For purposes of illustration, exemplary arrangements of components are provided and combinations, additions, rearrangements, and the like are contemplated in alternative embodiments of the present invention. Accordingly, the invention has been described with respect to exemplary embodiments, and those skilled in the art will recognize that many modifications are possible.

For example, the processes described herein may be implemented using hardware components, software components, and / or any combination thereof. The description and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. It will be understood by those skilled in the art that various changes and modifications can be made therein without departing from the broader spirit and scope of the invention as set forth in the claims and that the invention is intended to cover all modifications and equivalents falling within the scope of the following claims It will be obvious.

Claims (32)

A method for selecting a playback rate in a receiver that receives media to playout using a presentation element of the receiver,
The playout results in the media being consumed at the playback rate from the presentation buffer and the receiver is configured to select from a plurality of playback rates,
Monitoring the presentation buffer, the presentation buffer storing media data between at least the time at which the media data is received and the time at which the media data is consumed by the presentation element associated with the receiver;
Storing an indication of a buffer level, the buffer level corresponding to how much of the presentation buffer has been occupied by media data that has been received but has not yet been consumed by the presentation element;
Determining an estimated download rate;
Using a stored indication and an estimated download rate to calculate a target playback rate; And
And selecting from among the plurality of playback rates in accordance with the target playback rate.
A method for selecting a playback rate. The method according to claim 1,
Wherein the selected playback rate is a playback rate that is less than or equal to a predetermined multiplication of the estimated download rate and wherein the predetermined multiplication is an increasing function of the buffer level. 3. The method of claim 2,
Wherein the predetermined multiplication is an affine linear function of the playback duration of the media data of the presentation buffer. 3. The method of claim 2,
Wherein the predetermined multiplication is less than one when the buffer level of the presentation buffer is less than the threshold amount. 3. The method of claim 2,
Wherein the predetermined multiplication is greater than or equal to 1 when the presentation duration of the media data in the presentation buffer is greater than or equal to a predetermined maximum amount of presentation time. 3. The method of claim 2,
Wherein the predetermined multiplication is a piecewise linear function of the playback duration of the media data of the presentation buffer. The method according to claim 1,
Wherein the selected playback rate is a playback rate that is less than or equal to a predetermined multiplication of the estimated download rate and the predetermined multiplication is a function of increasing the number of bytes of media data in the presentation buffer Lt; / RTI &gt; The method according to claim 1,
Wherein the playback rate is selected as the largest available playback rate among a plurality of playback rates that is less than or equal to the proportional factor multiplied by the estimated download rate and the proportional factor is divided by an estimate of response time to rate changes And a function of increasing the playback duration of the media data in the presentation buffer. 9. The method of claim 8,
Wherein the estimation of the response time is an upper bound on the presentation time between switch points of the media data. 9. The method of claim 8,
Wherein the estimation of the reaction time is an average of a presentation time between switch points of the media data. Wherein the estimate of the reaction time is greater than or equal to a predetermined constant multiplied estimated round-trip time (&quot; ERTT &quot;). A method for selecting a playback rate in a receiver that receives media to playout using a presentation element of the receiver,
The playout results in the media being consumed at the playback rate from the presentation buffer and the receiver is configured to select from a plurality of playback rates,
Monitoring the presentation buffer, wherein the presentation buffer stores media data between at least the time at which the media data is received and the time at which the media data is consumed by the presentation element associated with the receiver;
Storing an indication of a buffer level, the buffer level corresponding to how much of the presentation buffer has been occupied by media data that has been received but has not yet been consumed by the presentation element;
Determining an allowed deviation of the buffer level;
Using the stored representation of the buffer level and the allowed deviation of the buffer level to calculate a target playback rate; And
And selecting from among the plurality of playback rates in accordance with the target playback rate.
A method for selecting a playback rate. 13. The method of claim 12,
Wherein the playback rate is selected based on an estimate of response times for high proportional factors, low proportional factors, estimated download rates, current playback rate, buffer level, and rate changes. Way. 14. The method of claim 13,
Wherein the high proportional and low proportional factors are both incremental functions of the playback duration of the media data of the presentation buffer divided by an estimate of the response time to the rate change and / A method for selecting a rate. 14. The method of claim 13,
Wherein the high proportional factor is greater than or equal to the low proportional factor. 14. The method of claim 13,
Wherein the playback rate is selected to be equal to the previous playback rate if the previous playback rate is between the low proportional factor times the estimated download rate and the high proportional factor times the download rate estimate Lt; / RTI &gt; 14. The method of claim 13,
Wherein the playback rate is selected to be the highest available rate multiplied by the high proportionality factor multiplied by the higher rate factor if the previous rate of play exceeds the high rate factor estimate, How to choose. 14. The method of claim 13,
Wherein the playback rate is selected to be the largest available play rate that is not greater than the low rate factor estimate multiplied by the low rate factor if the previous play rate is less than the low rate factor estimate multiplied by the download rate estimate Lt; / RTI &gt; A receiver that receives media to play out using a presentation element of a receiver and consumes media data at a playback rate,
A presentation interface for providing playback at one of a plurality of playback rates;
A presentation buffer storing the media data and being coupled to the presentation interface between at least the time at which the media data is received and the time at which the media data is consumed by the presentation element associated with the receiver;
Storage for variables relating to presentation buffer capacity, including an indication of a buffer level, said buffer level indicating how much of said presentation buffer has been occupied by media data that has been received but has not yet been consumed by said presentation element Corresponding;
An estimated download rate determiner for determining an estimated download rate;
And logic for arranging the requests according to the determined selected playback rate using the stored indication and the estimated download rate to calculate a target playback rate.
receiving set. 20. The method of claim 19,
Wherein the selected playback rate is a playback rate that is less than or equal to a predetermined multiplication of the estimated download rate and wherein the predetermined multiplication is an increasing function of the buffer level and wherein the predetermined multiplication is performed in the media of the presentation buffer Receiver is an affine linear function of the playback duration of the data. 21. The method of claim 20,
Wherein the predetermined multiplication is less than one when the buffer level of the presentation buffer is less than a threshold amount. 21. The method of claim 20,
Wherein the selected playback rate is a playback rate that is less than or equal to a predetermined multiplication of the estimated download rate and wherein the predetermined multiplication is an increasing function of the number of bytes of media data in the presentation buffer. 20. The method of claim 19,
Wherein the playback rate is selected as the largest available play rate among a plurality of playback rates that is less than or equal to the rate factor estimate multiplied by the download rate estimate and wherein the proportional factor is divided by an estimate of response time to rate changes And a function of increasing the playback duration of the media data of the presentation buffer. 24. The method of claim 23,
Wherein the estimate of the reaction time is an average over a presentation time between switch points of the media data or a presentation time between switch points of the media data. 24. The method of claim 23,
Wherein the estimate of the reaction time is greater than or equal to a predetermined constant times the estimated round-trip time (&quot; ERTT &quot;). 18. A non-transitory computer readable medium for execution by a processor of a receiver for playing out media to a presentation element of a receiver,
The playout results in the media being consumed at the playback rate from the presentation buffer and the receiver is configured to select from a plurality of playback rates,
The medium may further comprise:
Program code for monitoring a presentation buffer, the presentation buffer storing media data between at least the time at which the media data is received and the time at which the media data is consumed by the presentation element associated with the receiver;
Program code for storing an indication of a buffer level, the buffer level corresponding to how much of the presentation buffer has been occupied by media data that has been received but has not yet been consumed by the presentation element;
Program code for determining an allowed deviation of the buffer level;
Program code for using the stored representation of the buffer level and the allowed deviation of the buffer level to calculate a target playback rate; And
And program code for selecting among the plurality of playback rates in accordance with the target playback rate.
Including program code,
Non-transient computer readable medium. 27. The method of claim 26,
Further comprising program code for selecting a playback rate based on an estimate of response time for the current playback rate, the buffer level, and rate changes, a high rate factor, a low rate factor, a download rate estimate, Transient computer readable medium. 28. The method of claim 27,
Wherein the high proportional and low proportional factors are both incremental functions of the playback duration of the media data of the presentation buffer divided by an estimate of the response time to the rate change and / Transient computer readable medium. 28. The method of claim 27,
Wherein the high proportional argument is greater than or equal to the low proportional argument. 28. The method of claim 27,
Further comprising program code for comparing and selecting the playback rate to be equal to the previous playback rate if the previous playback rate is between the low proportional factor times the estimated download rate and the high proportional factor times the download rate estimate &Lt; / RTI &gt; 28. The method of claim 27,
Program code for comparing and selecting the playback rate such that the highest rate multiplied by the high rate factor multiplies the high rate rate multiplied by the high rate factor and is the largest available playback rate not greater than the estimated download rate &Lt; / RTI &gt; further comprising a computer readable medium. 28. The method of claim 27,
Program code for comparing and selecting the playback rate such that if the previous playback rate is equal to the low rate factor multiplied by the download rate estimate then the low rate factor is the largest available playback rate that is not greater than the download rate estimate &Lt; / RTI &gt;

Images